New APM Stats

[bondom34]

Anyone else checking these out? Pretty cool stuff, some new ones include adjusted points per shot, adjusted points per shot effect on teammates, Pair plus/minus, adjusted TOs, and more. A few interesting results as well.

Manu still is awesome, greatly positive influence on teammates PPS. Strangely Raymond Felton is up there too.

OKC is way up in adjusted PPS for individuals w/ Collison, Ibaka, Durant, and Westbrook all at +2 overall or better and Sefolosha at 1.9.

Lots of cool stuff!

[Doctor MJ]

Link?

[bondom34]

Link?

http://stats-for-the-nba.appspot.com/

[Doctor MJ]

Cool stuff. More access to individual factor regression data is something we need. The pair regression is a direction I've been wanting to see more of too.

[bondom34]

Cool stuff. More access to individual factor regression data is something we need. The pair regression is a direction I've been wanting to see more of too.

I thought the pairs were something really important, and really are insightful to me. Just watching OKC a lot then comparing, you can clearly see w/ the eye test value of a guy like Collison and how poor Perkins can be offensively in particular, and the pairings completely bear it out as well. It also showed for me Jeremy Lamb's strengths (he does play well particularly w/ Collison out there and they're up there in the pair ranks). The coach data was interesting, but I think there are some kinks to be worked out. Just lurking on APBR and it seems Engelmann knows this and is working on improvements.

[Doctor MJ]

Cool stuff. More access to individual factor regression data is something we need. The pair regression is a direction I've been wanting to see more of too.

I thought the pairs were something really important, and really are insightful to me. Just watching OKC a lot then comparing, you can clearly see w/ the eye test value of a guy like Collison and how poor Perkins can be offensively in particular, and the pairings completely bear it out as well. It also showed for me Jeremy Lamb's strengths (he does play well particularly w/ Collison out there and they're up there in the pair ranks). The coach data was interesting, but I think there are some kinks to be worked out. Just lurking on APBR and it seems Engelmann knows this and is working on improvements.

I think pair data is a good start toward evaluating synergy & redundancy in team set ups. Just really important stuff.

The coach data though, meh. It's actually not new, they started making that not long after APM came out , and I've just never been impressed. You can only do anything with it when there's a coaching change, and a coach goes through small enough number of such changes that you can just analyze that in your head.

[Chicago76]

I just looked over this. Interesting, but I'm a little puzzled by the interpretation of teammate effects. Using DRB teammate impact: is more really "better"? Robin Lopez is the worst defensive rebounding starting C in the league over the last 2 years (14.4 DRB rate). His teammates' collective DRB rate jumps 5.82. This is supposed to be good. Are his teammates' boards jumping because he isn't cannibalizing their DRB opportunities, or are they jumping because he's not a particularly great rebounder, so they're doing some of his work? I'm going with the latter. Tim Duncan has posted one of the best DRB rates in the league for bigs. His teammates' collective DRB declines by 6 when he's on the court. He might be "stealing" some boards, but if the team as a 5-man unit is posting strong DRB rates, I'm okay with that.

Rather than looking at teammate DRBs, TOVs, etc, which not just develop RAPM totals for the key factors for players based upon team output of these factors when the player is on the court? I care about team effects, not who is necessarily doing the work or who is stealing from whom.

-DRB, opp FTA/FGA, 3FGA/2FGA, opp eFG%, opp TS%, opp TOV forced, steal rates, assisted basket frequency
-ORB, FTA/FGA, 3FGA/2FGA, eFG%, TS%, TOV rates, assisted basket frequency

unlike pts/poss in consolidated RAPM, some of these things might not happen frequently enough to get a decent read on medium-minute players, but it would still be nice to see for players on the court enough to produce the requisite sample size.

[mysticbb]

I just looked over this. Interesting, but I'm a little puzzled by the interpretation of teammate effects. Using DRB teammate impact: is more really "better"? Robin Lopez is the worst defensive rebounding starting C in the league over the last 2 years (14.4 DRB rate). His teammates' collective DRB rate jumps 5.82. This is supposed to be good. Are his teammates' boards jumping because he isn't cannibalizing their DRB opportunities, or are they jumping because he's not a particularly great rebounder, so they're doing some of his work? I'm going with the latter. Tim Duncan has posted one of the best DRB rates in the league for bigs. His teammates' collective DRB declines by 6 when he's on the court. He might be "stealing" some boards, but if the team as a 5-man unit is posting strong DRB rates, I'm okay with that.

I completely agree with that. Engelmann had previously made a study based on the effect of a player on the team overall rebounding, which made much more sense. Identifying the player who helps achieving a better than league average rebounding result (and at that on the defensive end) would be much more telling. Just to illustrate the difference a bit: Nene was the top player in that study on the defensive end, improving the team defensive rebouding over average by about +2.2. In this current format, Nene is also considered a positive factor for his teammates, but overall he is rather down in the list. But the first study was actually really telling us something about the rebounding impact by an individual player, while this current version just tell us something about the collected rebounds. Engelmann considers the current version as better to describe the individual player, while for the game itself I consider the impact study clearly more valuable, because it is in agreement with results for the team when specific players were on the court. When we look at this list here (3800 min is about 1/4 of the overall possible playing time during that timestretch): http://bkref.com/tiny/7Krgm

I find that particular telling that Nene is leading the league here over that timespan, while you have to scroll down the list pretty far before one of the players who shared time with him on the court also appears. Given Nene's own rather low DRB%, we can dig deeper and find him to a guy being at the top level of moving under the own basket and boxing out, preventing the other team to get the offensive rebound. That is useful information. But Engelmann didn't like the fact that the values of the impact study barealy correlated with the individual DRB% (R² of about 0.13 when I tested it back in 2011). Now he has an approach which he deems to be closer to "describing the individual player".