How Well do NBA Win Shares Correlate to Actual Wins?

In an earlier Statistical Test on this here quality NBA publication, we took a look at Value Over Replacement Player (or, as it’s known in the Pace and Space offices, One Stat to Rule Them All) and found a strong correlation between VORP and team wins that can describe a team’s win total so accurately that if you have a reasonably good way to predict a roster’s VORP in advance, you’re more than halfway to beating the over/under game in Vegas.

But there’s another stat out there that claims to do the same thing, and they even went so far as to outright call it Win Shares.

And because I am nothing if not suggestible, when someone asks me on Twitter to do the math…

…then let’s do the math. Methodolgy will be exactly the same as in that VORP piece, so if you’ve read that one, I don’t need to explain the method again. If you haven’t read that one…go read it! I’ll wait.

(note that these totals are sourced from Basketball Reference and include win shares credited to a player during his tenure with the team to whose total he contributed.)

Now then.

Eastern Conference

(format here is team, wins, Win Shares. Commas are for obvious reasons for anyone who’s ever used Excel.)

MIL, 60, 62.5
TOR, 58, 56.2
PHI, 51, 48.8
BOS, 49, 52.1
IND, 48, 48.0
BKN, 42, 42.0
ORL, 42, 43.3
DET, 41, 42.3
CHA, 39, 38.5
MIA, 39, 40.6
WAS, 32, 34.5
ATL, 29, 27.8
CHI, 22, 21.5
CLE, 19, 19.5
NYK, 17, 20.8

Wow. Biggest deviation (shock!) was the Knicks, who managed to be worse than their projected win total based on win shares, and the Celtics (woeful underachievers) were second. The Pacers and Nets managed to hit their number bang on, while a couple of other teams were within a decimal place against an integer of hitting theirs on the nose.

Based on this, things are looking good for a correlation almost as good if not better than the one we got from VORP, but let’s bring the West into it.

Western Conference

GSW, 57, 56.8
DEN, 54, 50.6
POR, 53, 51.7
HOU, 53, 53.3
UTA, 50, 53.7
OKC, 49, 49.9
SAS, 48, 46.2
LAC, 48, 43.3
SAC, 39, 39.0
LAL, 37, 35.7
MIN, 36, 37.5
MEM, 33, 35.8
NOP, 33, 37.2
DAL, 33, 38.8
PHX, 19, 19.5

The West, as with the VORP test, was a lot noisier statistically, as there were tiers between the 2-5 seeds, the 6-8 seeds, the cluster fiasco between 9 and 14, and the Warriors and Suns standing alone as the best and the worst.

But broadly, win totals and Win Shares tracked, with Sacramento hitting their number bang on the nose and three other teams coming within a decimal point of them.

Putting the Data Together

There’s no need for a super-advanced calculation step like the VORP formula, since 1 WS = 1 win in theory.

In practice: The best-fit line is actually (WS*0.94)+3.14 rather than a simple y=x, but close enough. A team with zero Win Shares should actually go 3-79.

But the r-squared? A whopping 0.963. Even more reliable than the VORP calculation, and the VORP calculation was more or less bang-on.

Conclusion

Let’s circle back to Zach’s original question about how Win Shares correlate to NBA champions.

Because the answer is simple: The overwhelming majority of the time, the answer is exactly the way wins and losses on the actual basketball court do. Statistically, they’re pretty close to one and the same.

This is also a fantastic endorsement of Win Shares as a stat. When it does its job this well, you know you can rely on it in an argument.