Two-Man Combinations

[eminence]

Basketball Reference has a stats table titled Regular Season: 2-Man Combinations, but only has it for the top 20 minutes combinations for each team. Does anyone know of any places where I could get a list of all 2-man combinations for a team? Am also looking for total #'s, not per100 #'s (though those can be estimated if need be). Really hoping I don't have to figure out how to mine it myself

End goal is to find Net Pts for 2-man combos and will then build some networks based off of that, looking at how it compares to RAPM.

[eminence]

Nvm, found it dating back to 2001, woohoo! It's on basketball reference in their lineup finder section.

To summarize what I'm doing: I'm graphing each team, with the players as the nodes and the 2-man combos as the edges, with the 2-man total +/- being the weight of the edge. Then finding the degree of each player. Hoping to find an alternative impact-stat to (r)apm that doesn't require adjusting for minutes while using a bit less computational effort and a simpler process (at least for me, the statisticians might disagree!).

Early thoughts (did Jazz/Warriors/Clippers from '15-'16 as a test, regular season and playoffs).

Making the graphs is pretty simple, and I'm thinking it does a pretty good job of ranking players in order of importance to their team(though not completely the order I expected - which is probably a good thing). Main problem I've run into is that my edges can't have negative values with the functions I'm using (if a two man combo has a -90 net I have to add 91 to all the edges so I can still use them all when graphing). This makes it pretty much impossible to compare values across teams, I was imagining I could just normalize them for each team and then compare those normalized values, something to think on.

Also have doubts it really says much more than net +/- for each player for the season (though there are some differences). After I've run some more teams I'd like to see how it compares to other Box-score/Tracking/Impact-stats by doing some regressions.

[eminence]

Early returns, fixed the bit with not being able to standardize the values. Jazz/Warriors/Spurs/Clippers done so far.

So the top 10 "most important to their teams?" from those 4 teams for 2015-2016 (RS and PO).
1. CP3 - 2.34
2. Draymond - 2.08
3. Kawhi - 1.97
4. Curry - 1.94
5. Redick - 1.86
6. Jordan - 1.84
7. Gobert - 1.74
8. Hood - 1.51
9. Klay - 1.45
10. Mack - 1.40

Edit -Couple small changes (removed the super low minutes guys from the standardization step, <500 mins, what do y'all think a good number for that would be?). Will have to do the Sixers tomorrow morning, thought just occurred to me that all of their top minutes guys will automatically wind up at the bottom, which is not what we're going for, we'll see.

With low low minutes guys removed Top 10:
1. CP3 - 1.96
2. Kawhi - 1.79
3. Draymond - 1.79
4. Curry - 1.65
5. Redick - 1.51
6. Jordan - 1.49
7. Gobert - 1.45
8. Hood - 1.24
9. Klay - 1.16
10. Mack 1.13

Moved the lower minutes bound to 1,000 minutes for players included in standardizing their contributions to the team, and added in the Sixers. With the minute change CP3 drops and Kawhi rises up. Sixers didn't flip it around like I worried they might, I think setting the minutes cut off high enough should avoid that (1,000 seems like a pretty good number, about the top 10 guys for each team).
1. Kawhi - 1.95
2. CP3 - 1.55
3. Covington - 1.54
4. Draymond - 1.53
5. Curry - 1.39
6. McConnell - 1.36
7. Gobert - 1.30
8. Redick - 1.14
9. Jordan - 1.12
10. Hood - 1.11

One more different thing to try for now, this time only including two man lineups where both players played 1000+ minutes for the team(next try will be with 2-man groups that played together for x minutes).
1. Kawhi - 1.88
2. CP3 - 1.48
3. Draymond - 1.47
4. Covington - 1.42
5. Curry - 1.36
6. Gobert - 1.29
7. Redick - 1.22
8. Ish - 1.15
9. Jordan - 1.12
10. Hood - 1.00