Well, it's back to the grind this week, and though I've mainly been working on the fantasy game (to try to get it out well before the playoffs begin), I've also added a small tweak to the player listings. In trying to accurately decide what position a particular player plays, I've taken some information from 82games.com. On that site, they show each player's position percentage played. For example, through the year, Tracy McGrady has played 38% of the Rockets' available minutes at shooting guard, and 25% of the available minutes at small forward. The page also shows McGrady using "0%" of the available minutes at power forward.
To calculate a particular player's position adjustment, I first add a tweak factor of 0.5% to each number. I have found that this does a very good job of reaching 100% for each position on the team. This leaves McGrady's percentages at 38.5% SG, 25.5% SF and 0.5% PF. I extrapolate those numbers to 100%, so for McGrady it would be 59.7% SG, 39.5% SF and .8% PF. Then I take the position adjustments found by regression in The Wages of Wins and calculate a position adjustment. (Scroll down to "Calculating PAWS").
Tracy McGrady
SG 59.7% .128 .076416
SF 39.5% .152 .06004
PF 0.8% .215 .00172
McGrady's new position adjustment is .138, which I think accurately reflects that he plays both shooting guard and small forward, but plays more minutes at shooting guard.
I have shared these thoughts with David Berri, and he seems to be receptive into this way of getting a player's position adjustment. The key is making sure that this automated method gives results that are close enough to the manual method of allocating out a team's minutes by hand. If any of you notice any results that seem strange to you, please leave a comment on this posting, and I'll take a look at them. It's possible that my scripts have missed a player here or there, or that the results are off. But hopefully this will be a good step to making this site closer to representing the Wins Produced formula accurately, so I can positively portray a particular player's position played plus produce a perfecter portrayal of a player's per-game production. Perhaps.
Subscribe to:
Post Comments (Atom)
10 comments:
shawn marion's Wins Produced (.385) is a lot higher than the one DBerri recently posted (.379).
sorry, that should be .279 instead of .379
Yes, I noticed that one too. I wonder if DBerri is calculating Marion as a pure power forward. 82games.com has him split between small forward and power forward, with a slight lean toward small forward.
The space between SF and PF is definitely the most variable spot. With the position adjustment of a PF being .215 and an SF being .152, there is a huge margin for error between those positions.
This seems somewhat troubling, especially in small samples where it could be affected by an outside force. For example, if Boris Diaw gets injured, Marion will probably play more PF. Does that mean he needs to have better stats to have a similar WP48? Yes. Is that right? I'm not sure. However, I have a feeling in larger samples, that sort of thing would even out.
As far as Marion goes, I tend to think he is one of the most valuable players in the league, and I like the higher number. But I'm not sure what DBerri would have to say about that.
I have Shawn Marion at power forward and Boris Diaw splitting his time at power forward and small forward.
Even if Diaw played all his minutes at power forward, though, Marion would still have to play the majority of his minutes at the four spot. At least, I don't see who else on the roster could be playing power forward.
After 72 games the Suns had 3501 minutes to allocate at each position. At center you have Stoudemire and Thomas. At power forward they have Marion and Diaw. After these four, who else on the roster can play in the frontcourt (James Jones, Raja Bell?)
All that being said, I am very hopeful we can intergrate the 82games positions into the calculation of Wins Produced. It certainly would make life easier.
Here is the breakdown of PF minutes from 82games.com:
Pat Burke 1%
Sean Marks 0%
Kurt Thomas 10%
Boris Diaw 49%
Shawn Marion 34%
Jumaine Jones 1%
James Jones 3%
I've written to 82games, asking how they determine the different position percentages. I'll post once they reply.
I looked at 82games.com. They have Diaw playing mostly power forward (and a bit of center). So I re-did my analysis, moving Diaw to power forward (entirely) and having Marion playing about half his minutes at small forward. When I do this, Marion's WP48 rises to 0.322.
The difference now is that WP48 adjusts for the opponent's field goals, opponent's turnovers (that are not steals, and team rebounds. Our simple PAWSmin model does not. For most teams and player, as I have noted before, this makes very little difference. For the Suns, though, it does make a bit more of a difference.
About positions... it looks to me that 82games.com does allocate minutes as I do, in the sense that they require that a team always employ a center, power forward, small forward, shooting guard, and center. Hopefully when the season is over we can use 82games.com position data in calculating WP48 and PAWSmin for each player.
jchan, could you add a filter for statistically significant minutes? if you try to sort games for each player on winscore to figure out what the best individual games were this season, you have to go through at least four pages because the first few are full of statistically insignifant performances (players who played for 1-9 minutes)
Iverson's best game appears twice in the stats...
Thanks for spotting that. Looks like I had a bunch of duplicate data for that day. Bad luck for Iverson, his approx. WP48 went from .115 to .102. Ouch.
Interesting story you got here. I'd like to read more concerning this topic.
BTW look at the design I've made myself High class escort
Post a Comment