Statistical Test: Correlating VORP and NBA Team Wins

Back in May, I examined the contributions of “replacement-level” NBA players, those guys with 0.1 either side of zero VORP, and in the process tried to figure out what an entire team of those guys or their presumed G-League “replacement player” equivalents would do over the course of a season.

And in that effort, I took a six-team sample size from a cross-section of NBA competence to determine a formula: Wins (per 82 games) = (2.15*aggregate VORP)+19.

But it occurs to me that I can do better than just taking a sample, shoehorning a little back-of-the-envelope math into it, and then trying to play that formula like it’s gospel truth.

Instead, let’s bring the other 24 NBA teams in 2018-19 into it, aggregating their VORP and seeing what a little actual line-of-best-fit analysis brings to the table. Plot wins on the Y axis, VORP on the X axis, and let the R-squared shot fly and see if the correlation is nothing but the bottom of the net or Josh Smith’s shot chart accuracy wise.

Let’s go top-to-bottom by wins in each conference because using the 2018-19 regular-season standings is a great way to prevent things like “forgot a team”. This is, by necessity, going to be math-heavy stuff.

Data source: Basketball Reference, using player games played for the team under discussion (available on the team’s season summary page.)

Eastern Conference

MIL: 60-22, 19.2 team VORP
TOR: 58-24, 16.5
PHI: 51-31, 12.5
BOS: 49-33, 14.9
IND: 48-34, 13.5
BKN: 42-40, 9.6
ORL: 42-40, 10.4
DET: 41-41, 9.6
CHA: 39-43, 8.6
MIA: 39-43, 9.5
WAS: 32-50, 6.2
ATL: 29-53, 2.9
CHI: 22-60, -0.1
CLE: 19-63, -2.0
NYK: 17-65, -0.9

You’ll notice already that the trendline isn’t perfect, and that some clear over- and under-achievers emerge. Philly and Boston up top, Cleveland in the cellar…and no surprises in either instance when you consider Philly being largely thought overrated, Boston’s chemistry being a mess, and the Cavs using a lot of Collin Sexton because they weren’t afraid to stink while he learns on the court. But a fair correlation even before putting it into chart form.

Onward to the West.

Western Conference

GSW: 57-25, 17.6
DEN: 54-28, 15.4
POR: 53-29, 15.5
HOU: 53-29, 15.9
UTA: 50-32, 16.4
OKC: 49-33, 14.2
SAS: 48-34, 12.2
LAC: 48-34, 11.1
SAC: 39-43, 9.2
LAL: 37-45, 8.5
MIN: 36-46, 9.0
MEM: 33-49, 7.2
NOP: 33-49, 8.6
DAL: 33-49, 9.1
PHX: 19-63, -0.2

A weird clustering of scores that don’t make perfect sense at three wins either side of 36, but that was pretty much the West “race” below playoff level all season, wasn’t it?

Now to do some more advanced math.

So How Relevant Is the Data?

Well, for one thing, my back-of-the-envelope calculation was in the ballpark. The true win expectation is 20.34+(2.07*VORP) for the entire league.

And the r-squared? A whopping 0.953. These numbers, by and large, hew VERY close to the line.

Plus, we get one huge takeaway from this as well. The average VORP for a 41-win team should be 10. We’re talking 10.00 to two decimal places (and a repeating three from the third place on) precise.

We’ll have some more fun with this data set (and throw in a couple from previous seasons) as the summer goes on. But it’s nice to see that expanding the sample size from six (in that original article about replacement player value) to 30 (the entire league) only fine-tuned rather than blew up the data set. It also gives us an easy way to determine who over- or underachieved last year; just plug their wins and VORP into the formula and see how far off they were one way or the other.

But mostly, what we’ve proven is that Sheed’s Law is immutable. Ball don’t lie…especially when you use Value Over Replacement Player to measure ball.