Win Shares and Trades (321 views)

[eric]

So I post Win Shares every year, and Bankz (I believe) wondered whether a 15 WS player on one team would generate 15 WS on some other team. This is a useful thing to know, and so I looked into it. I took the default 2005 league and selected three players: Raef LaFrentz, Nick Collison, Kevin Garnett. They were each the first power forward on their teams, backed up by a player who also backed up the center, and I set all their strategies the same (fast/often/normal/balanced). I measured 80 seasons of Raef on Boston and Nick on Seattle, 80 seasons with them on the opposite team, 80 with Raef/KG, 80 with KG/Raef. Here were the results:

Raef had 5.9 WS in 2371 minutes on Boston (.119), 4.9 in 2122 on Seattle (.111).
Nick had 4.1 / 2338 = .084 on Seattle, 4.2 / 2403 = .084 on Boston.
Boston went from 34.2 to 31.9 Win Shares = 2.3 compared to the player change of 1.7
Seattle went from 45.6 to 49.2 = 3.6 vs 0.8

Raef had 5.9 / 2395 = .118 on Boston, 4.8 / 2131 = .108 on Minnesota.
Kevin had 18.4 / 3049 = .290 on Minnesota, 17.7 / 2981 = .285 on Boston.
Boston went from 35.2 to 50.3 = 15.1 vs 11.8
Minnesota went from 55.2 to 40.9 = 14.3 vs 13.6

The players stayed remarkably similar even when traded on a per minute basis. It was interesting to see how the playing time changed: pretty significantly, and not reciprocally. Raef played less minutes in Minny, suggesting his backup was better or the rest of the team defense didn't put him in position to foul as much or something, but Garnett played less minutes when he went to Boston, so none of those explanations makes sense. It is something we can control for, though. We know the wins and minutes of the starters. We don't know the backups' WS/48, but we can calculate it and see if it's plausible.

Minnesota - Kevin = 55.2 - 18.4 = 36.8
+ Raef = 36.8 + 4.8 = 41.6
actual - predicted = 40.9 - 41.6 = -0.7
/ minutes gap * 48 = -0.7 / (3049 - 2131) * 48 = -.037 WS/48

Boston - Raef = 35.2 - 5.9 = 29.3
+ Kevin = 29.3 + 17.7 = 47.0
actual - predicted = 50.3 - 47.0 = 3.3
/ minutes gap * 48 = +3.3 / (2981 - 2395) * 48 = +.270 WS/48

Neither is super plausible. Backups are usually in the .050 to .100 range. There's always noise involved, and noise of ± 2 doesn't seem too bad, and I'm comfortable saying that Win Shares turn out to be a pretty good predictor of the Win Shares a player will generate for a different team. I also kept track of regular wins and Pythagorean wins to see if there was anything going on there.
team: shares / reg / pyth
Boston: -2.3 / -1.1 / -2.9
Seattle: 3.6 / 2.4 / 2.2
Boston: 15.1 / 13.2 / 19.2
Minnesota: -14.3 / -12.4 / -17.3

Bottom line, the arithmetic is straightforward no matter how you're measuring wins. Dropping a 5 WS player for a 15 WS player is going to get you about 10 wins.

[Bankz]

So a player on a worse team will produce a lesser win share it seems and vice versa. It may not be a large gap but it does change.

Or am I reading this wrong eric

[eric]

Dec 30, 2015 15:23:19 GMT -6 Bankz said:

So a player on a worse team will produce a lesser win share it seems and vice versa. It may not be a large gap but it does change.

No, I don't think so. Seattle was better than Boston but both Raef and Nick produced slightly less in Seattle. Raef also produced slightly less going from the worse Boston team to the better Minnesota team, but Kevin produced slightly more on the better Minnesota team.

It looks to me like players produce the same per minute wherever they go. The bigger possible change is how many minutes they get, but even that is pretty small.

[MistleTacoe]

Thanks Eric