Looking for Raw Team Data in On-Off Scenarios

[Chicago76]

I'm trying to test a theory re: penetrators, post men, and spacers and their impact on assisted buckets and FG%.

To do so, I'm trying to find raw on-off data that 82games does not seem to have available. What I'm looking for is the following for a teams when player X is on and off the court:

FGM-FGA
3M-3A
FTM-FTA
AST
TOV
ORB
TRB

Without diving too much into things, what I'm trying to identify is the indirect impact of a player's presence on the court in terms of assist creation. Using Brad Miller as an example from 2005-06, 82games can give me a rough estimate that, when Miller was on the court per 100 poss:

The Kings made ~45.2 FG, 28.5 of which were assisted.
Miller made ~7.6FG, 5.6 of which were assisted.
Which means the rest of his teammates were assisted on 60.8% of their baskets, which is higher than when Miller was off the court (55% assisted baskets).

Attributing all of the excess to Miller = +2.1 additional assists per/100 when he is on the court. Dividing the remaining baseline 55% assisted basket rate across non-shooting teammates is another 5.2 ast/100 for Miller.

"Adjusted assists" = 7.3 ast/100 poss. His actual assists were 6.7/100 poss. A couple more:
Nash (last year) 13.3 vs. 17.7
Rose (this year) 9.2 vs. 11.0

What I'm after is an "assist shares" metric that might allow me to tease out the relationship between spacers, big men, passers, and penetrators and how they all kind of work together to promote assisted buckets.

The issue w/ 82games is that they use only eFG%, which doesn't allow me to estimate actual made FGs. Without this, I'm stuck assuming that the distribution of made 2s and made 3s is the same regardless of who is on the court.

[SideshowBob]

I think this might be in the realm of what you're looking for

NBA in general:
http://www.basketball-reference.com/play-index/plus/plus_minus_finder.cgi

For example, Miami's players sorted by Assists per100 possessions when player is on the floor

http://bkref.com/tiny/Jwkvz

[Chicago76]

I think this might be in the realm of what you're looking for

NBA in general:
http://www.basketball-reference.com/play-index/plus/plus_minus_finder.cgi

For example, Miami's players sorted by Assists per100 possessions when player is on the floor

http://bkref.com/tiny/Jwkvz

For some reason, I thought this newer b-r feature only contained the net team +/- by statistical category. Thank you. Thank you. Thank you.

[SideshowBob]

Yeah, I was surprised and delighted at how comprehensive this new feature was.

Your welcome I'm looking forward to reading about your findings!

[Chicago76]

Well, I took a first crack at this, and while things generally make a bit of sense, there is a lot of noise due to teammate interaction.

What I did was compare the the % of field goals that were assisted when player X was off the court to the % of player X's teammates FGs that were assisted when he was on the court. Looking only at teammates when a player was on the court is an important distinction.

If a player is good at creating his own shot (like a Rose), then the % of FGs that are assisted for the entire 5-man unit should be (and is) lower than when he is out of the game. He's a huge part of the offense, and he's creating on his own, so there will be fewer assists to go around. Similarly, a big man that relies upon being fed in the post will drive up the % of FGs that are assisted when he is on the court. He's not acting as a facilitator by rather as a consumer of possessions. The top ten is PG heavy as, you would expect.

Here are the first 16 of three groups (1000 min+ in 2010/11). Groups are PGs, SF/SGs, and PFs/Cs

PGs: Nash (+20.5%), Paul (+20.4%), Rondo (+17.3%), Calderon (+13.2%), Parker (+12.9%), Augustin (+11.8%), Kidd (+11.2%), J. Nelson (+10.9%), Westbrook (+10.6%), Holiday (+10.5%), Wall (+9.4%), A. Miller (+8.1%), CJ Watson (+8.1%), Rose (+7.6%), Collison (+7.6%), Curry (+7.2%).

SG/SFs: J Johnson (+10.5%), Ginobili (+8.8%), Pierce (+8.1%), James (+8.0%), Allen (+7.7%), Iguodala (+7.2%), Grant Hill (+6.9%), Wade (+6.1%), Ellis (+5.5%), R Hamilton (+4.7%), E Gordon (+4.7%), Kevin Martin (+4.6%), Roy (+4.5%), Dunleavey (+3.9%), S. Jackson (+3.7%), Ariza (+3.5%).

PFs/Cs: Aldridge (+10.8%), Horford (+9.7%), J Smith (+9.5%), Griffin (+7.3%), Diaw (+6.8%), D. Lee (+5.3%), Milicic (+4.9%), Hawes (+4.8%), Blatche (+4.5%), Chandler (+4.3%), Ilgauskas (+4.3%), Kenyon Martin (+4.0%), Duncan (+3.9%), Hilario (+3.6%), D. Howard (+3.5%), McRoberts (+3.3%).

There seems to be a ton of teammate interaction in these numbers. Examples:

Rose: positive contribution. Deng: very negative contribution of about -8%--likely due to being on the floor so much w/ Rose, who tends to create unassisted baskets.

Horford/Smith/Johson are all solid, but there appears to be a lot of the excess due to the trio playing together.

Aldridge was a major surprise. Probably largely due to a poor off court sample given his high minutes+the cupboard being pretty bare in Portland.

If I was technical enough (and smart enough), I'd probably take the raw numbers and figure out how to run lineups to control for player subsition patterns in a manner that is similar to taking raw APM and translating it to standardized.