Adjusted Impact Rating- Top down metric, just noticed

[Crow]

Link found on Evan's small forum. http://adjustedimpactrating.weebly.com/

2015 data: http://adjustedimpactrating.weebly.com/ ... ng-15.html

[j_oxford]

Hello, I'm the creator of Adjusted Impact Rating. I'd be glad to answer any questions on it. I think it's strongest attribute is identifying role players (starting or off the bench) who do a great job of keeping the team playing at it's highest level. Thabo Sefolosha is my favorite example, primarily due to how the Hawks declined after his unfortunate off-court injury. They obviously had other injury problems later in the playoffs, but their play was faltering going into the postseason. Tony Allen is another guy who doesn't get as much credit as he probably should. Memphis played at a much higher level with him on the court as opposed to the normal starters. He shared time with all of them of course, but his ability to come off the bench and elevate the team to that degree is astounding, and probably underrated.

[Crow]

Glad your account got activated. Thanks to Daniel.

My initial reaction is to somewhat agree with you that the metric is probably most helpful when used as a rating for a team's option over, perhaps, a league rating.

Any interest in checking the correlation of these values to raw plus minus, RPM and one or more boxscore metric? I am guessing higher correlation to raw plus minus than RPM and am curious about correlation to BPM, EZPM, WS, DRE at Nylon Calculus, EWins here by Mike G.

Did any of the weights of the elements of AIR surprise you?

Your approach is notably different than others. One worth checking. Another lense.

[j_oxford]

Glad your account got activated. Thanks to Daniel.

My initial reaction is to somewhat agree with you that the metric is probably most helpful when used as a rating for a team's option over, perhaps a league rating.

Any interest in checking the correlation of these values to raw plus minus, RPM and one or more boxscore metric? I am guessing higher correlation to raw plus minus than RPM and am curious about correlation to BPM, EZPM, WS, DRE at Nylon Calculus, EWins here by Mike G.

Did any of the weights of the elements of AIR surprise you?

Your approach is notably different than others. One worth checking. Another lense.

I haven't done an actual correlation study yet, but I definitely wouldn't be opposed to doing one.

Offensive rebounding being more valuable than defensive rebounding was a surprise, even if the difference wasn't large. The gap between the rebounding coefficients was proportional to the gap between turnover coefficients, but the gap between TS% was lower than would have been expected given the other gaps.

[Crow]

Some further probing on ts% issue might be useful / important.

fyi, if interested in bb history, also check out http://www.apbr.org/forum/

[j_oxford]

Some further probing on ts% issue might be useful / important.

fyi, if interested in bb history, also check out http://www.apbr.org/forum/

I expected a gap in Rebounding %, but the opposite way.

I had no prior assumptions about whether offense or defense would be more important for TOV% and TS%, but the model indicated limiting your own turnovers is more important than forcing the other team to commit them (this hurts as a Rockets fan, but hopefully Lawson can remedy that a bit), and similarly increasing TS% is more important than decreasing opponents TS%. It could be a sample size issue (the trainer set was only 1 season, 2014), or it could be that those things are actually more important. more research is definitely necessary.

[Crow]

Have you tried using AIR for lineup construction in your NBA analysis or coaching venture?

Do you think AIR has advantages and / or disadvantages for this use compared to RPM?

Would you consider trying AIR splits for different lineups or lineup "types" or by presumed position played? Is AIR boxscore based now? Interest in eventually taking it to play by play or lineup level?

If you are super gungho, perhaps consider making AIR splits similar to some of what I described in my first comment in this thread viewtopic.php?f=2&t=8964

[j_oxford]

Have you tried using AIR for lineup construction in your NBA analysis or coaching venture?

Do you think AIR has advantages and / or disadvantages for this use compared to RPM?

Would you consider trying AIR splits for different lineups or lineup "types" or by presumed position played? Is AIR boxscore based now? Interest in eventually taking it to play by play or lineup level?

I've done writeups on teams using AIR and Four Factor ratings in applications with those respective teams, one of which led to an interview (obviously didn't get the job lol), and suggested lineups to use/avoid. I'd like to eventually do separate regressions for groupings of players based on role. Things like 3&D, Stretch big, paint protector, distributing guard etc. I got the idea from basketball analytics by Shea and Baker, and I think it would result in better ratings. I just need a much larger sample size.

AIR isn't affected by boxscore metrics at all, Impact Rating is only 6 variables, all of which are at lineup level. which is the primary difference between it some of the other +/- metrics that give most credit based on boxscores. Again I haven't done a correlation study yet, but just from the eye-test it seems like AIR and RPM are similar in who is positive/negative, although to different degrees.

singling out AIR for certain lineups could certainly be done.

[Crow]

MCW's rating swung 9 pts moving from Sixers to Bucks. Did he and how he was used get that much better or were both ratings more about the other 4 guys playing with him? Or some of both? Checking his boxscore stats would help in assessing. BPM suggests he improved about 2 pts per 100 possessions. Raw on/off changed by about 10, equal to AIR change. Just one case but is AIR dominated by raw on/off despite some efforts to avoid?

[j_oxford]

The Y-input in the Impact Rating regression is Net on-court RTG (ORTG-DRTG), so it is designed to mirror it. The Impact Rating equation is a line of best fit describing Net on RTG given the 6 variables. It's possible I need to adjust the weighting on the teammate adjustment a little bit, but through trial and error I know its already really close to the appropriate weight.

[j_oxford]

As for MCW, I'd say the fit was the biggest part that helped him. Can't see him making that drastic of an improvement himself midway through the season. Teammates are certainly part of it, but the team played really well with him on the court, and he isn't the only one who got to play with those teammates.

[Crow]

Kanter team to team change:
AIR improved by 2
BPM improved by 5
Raw on/off change, none.

[Crow]

Should the teammate strength adjustment be weighted by time on court with player being weighted and the performance rating of the pair instead of or in addition to just the AIR of each of the court partners? Is using the average AIR the key difference between AIR and RAPM (setting aside regularization and priors in RPM)?

[Crow]

Mozgov team to team change
AIR almost plus 10
BPM 2.5
Raw on / off about 10

[Crow]

Jerebko
AIR plus 1.5 and plus 7 for a weighted avg of about plus 4
RPM plus 2.6
BPM plus 1
weight raw on/off about -2.5

Kobe
AIR -5.5
RPM -2.6
BPM about 0
raw on/off -5


Hard to reconcile the different numbers but perhaps with careful study of the results, metrics and contexts it can yield insight or at least theories. Maybe some needed to be blended, perhaps some left out. Need more systematic analysis.