What Do Analytics Say About 2016-17 NBA MVP Race?

Can analytics provide a nearly-objective perspective when deciding who should win this season’s NBA MVP?

If you’re curious about understanding the SteadyLosing MVP model that I am using for this latest update, I recommend reading Part I and Part II of my unanimous MVP discussion. This late-season 2017 NBA MVP update is the third part if you’re counting.

It’s quite clear that we won’t be seeing another unanimous MVP this season. It would be extremely difficult for one of the candidates to separating himself from the pack over the last 11-12 games of the year considering how tight the race is. Unlike any other season, the 2017 NBA MVP race has been more controversial than ever before, and for good reason. A good handful of players have legitimate cases with very few flaws. I will be using the SteadyLosing MVP model to get a barometer about how media votes may be allocated and will focus on a likely winning candidate.

Once again, here are the independent variables for my initial research:

The model is based upon the shares of maximum MVP points from 144 NBA MVP candidates over the last 10 NBA seasons.

• Note 1: Win Shares, which is a formula originally from MLB sabermetrics, seems to be quite a contentious statistic across NBA Twitter. That’s very understandable. Much like PER, many formulas in sports analytics attempt to measure player production in order to enhance our understanding and obviate subjective sources. Of course, with that, there are shortcomings. But for the sake of this MVP model, it may help to think of Win Shares as the expression of the marginal impact that a player has on his team’s offense and defense, while on the court. Furthermore, because James Harden spearheads an offensive attack that is actually one of the most potent the NBA has ever seen, Offensive Win Shares generously rewards his production. If James Harden was a redoubtable individual offensive player and the Houston Rockets merely had a decent offensive attack, with respect to the League average, his Offensive Win Shares total would be lukewarm. Conceptually, it’s not much deeper than this. Win Shares don’t claim to be the cornerstone of NBA metrics, but it’s not entirely useless
• Note 2: While some of the coefficients may seem confounding, the soundness of the model exists, and the results can spur solid discussion about the reasonableness of an MVP selection. (This is discussed further in Part II.) With greater access to more research stats and a ridge regression of the coefficients, the explanatory power of this model could increase.

Alright, who is the NBA MVP?

In order to be considered the season’s MVP, a player needs to have the greatest predicted share of maximum MVP points. The player first receives a raw, unadjusted score given the sumproduct of the coefficients and variable inputs. Thereafter, I recalibrate the unadjusted values in order to account for the amount of credentialed media members who will vote. For simplicity’s sake, let’s assume that 131 votes are collected, like last season.

Here are the results according to the SteadyLosing MVP model:

*Green box indicates that Kevin Durant and Kyle Lowry would’ve likely had higher consideration, but near-season ending injuries inhibited their statistical production.

James Harden is projected to be the 2016-17 NBA MVP!

Stephen Curry thinks that James Harden is the MVP, and regression agrees in this case. The graphic projects that James Harden will receive 708 of 1,310 maximum available points.

It’s also important to note that the distributions from the recorded share tallies of the model are often clustered. Just take a quick look at the predicted shares during last season:

Curry’s adjusted value of .857 was the greatest share total of any of the 144 players during any of the 10 seasons that I studied.

Next time, I will attempt to redistribute the unadjusted shares so that the allotments better reflect the skewness of the real NBA voting trend.

Additional Commentary:

• If you prefer to avoid James Harden as the MVP simply because of his turnover tally, then at least pace-adjust those numbers to give a fair comparison to his competitors. Nevertheless, Houston’s win total and offensive outburst have greatly exceeded expectations and he is a big reason for that.

• OKC fans will be pleased to hear that the recent stretch the Thunder have put together makes Westbrook’s MVP case more favorable than ever before, even according to analytics! However, It is possible that the MVP race could be even tighter than the model suggests because of the affinity that certain voters have for triple-doubles (For which I could start another rant about mathematical induction, the obsession with the tens place over the ones place, and how the fact that James Harden is nearly averaging a triple-double makes “But he’s averaging a triple double!” a moot point).
• Additionally, the narrative that Russell’s teammates are woeful could enhance his status among the voting community as well. The “without player X, we’d be 2-80” has been applied to almost every top MVP candidate there is, and I’d consider it a little too subjective of a stance to significantly elevate an MVP case.
• For weeks, as I provided small updates given the model, Westbrook had been mired between 4th and 6th, primarily because of his team’s win pace. It’s likely that Lebron’s hiatuses & Durant’s injury helped ignite a huge comeback in the model.

• Speaking of Kevin Durant, injuries essentially undercut his MVP portfolio. He was statistically sound for the majority of the year and the Warriors have struggled in his absence. However, a “Kevin Durant for first-place MVP” platform can’t really be helped too much at this point.
• Kyle Lowry didn’t get love from the model either because of his absence. In the future, I may consider testing a binary independent variable along the lines of “Was There A Stretch of Incompetence Directly Following Player Injury? (Yes or No)”.

• LeBron James can be the league’s best player and still not win an award that is designated for the most impressive production over the course of a single season. He has his eyes on a bigger picture but still impresses during the season. He’s always excellent, and his 3-point accuracy has recently skyrocketed:

man, good luck if LeBron is gonna be shooting like this in the playoffs pic.twitter.com/3QgmjKcLFR

— Jordan Zirm (@clevezirm) March 20, 2017

• Undoubtedly, Kawhi Leonard has shown that he is a superstar in the league. He’s simply becoming more and more proficient on offense while maintaining a unique perimeter defensive prowess (despite the confounding results of net rating and DBPM which have been examined by Bo Schwartz Madsen of Nylon Calculus).

• Relative to last year, Stephen Curry has been a little underwhelming. Just a little. However, he is crucial to his team’s win total, as his league-leading net rating of 16.2 portends. And as discussed in Part II, media members tend to value Team Wins considerably more than many other considerations.

• Let’s show love to Rudy Gobert who may become the Defensive Player of the Year. If it weren’t for a certain premier perimeter defender and rim protector (Draymond Green) who is sporting a 94.6 DRTG, Rudy would be a clear favorite. Nevertheless, Rudy Gobert is on pace for an eye-opening amount of Win Shares which shouldn’t be undervalued.
• Isaiah Thomas, yclept “King of the Fourth Quarter”, has put together a marvelous year, but an analytically-driven model hurts his stock because of his defensive limitations–which is largely due to size rather than effort. Nevertheless, he’ll get consideration from the committee as well.
• Why is John Wall so low? Could it be because of their recent slump which disturbed their win pace? Or, maybe his EFG% (.473) which ranks far below other superstars’ highlight some holes in his season campaign.

But, of course, this race is more jam-packed than usual. All we ask for is that committee members have persuasive justifications for their important decision.