A new top-down metric I've designed, the blog can be found at adjustedimpactrating.weebly.com Here are the Rockets ratings for the 13-14 season: Aaron Brooks: -2.364 Chandler Parsons: 2.789 Donatas Motiejunas: .569 Dwight Howard: 4.203 Francisco Garcia: .74 James Harden: 4.407 Jeremy Lin: 1.164 Omer Asik: .52 Omri Casspi: 3.171 Patrick Beverley: 5.897 Terrence Jones: .67 A complete description of the stat, along with the ratings of the rest of the league can be found at the above link. Go Rockets!
It is similar to a 4 factors model that uses eFG%, REB%, TOV% and FT/FGA. I used TS% instead of eFG% and FT/FGA because I think there is value in fouling the right free throw shooters, and that isn't expressed without using TS%. I didn't use points as an input because I have Net ONRating (ORTG-DRTG while player is on court) as an output, and there isn't a way to measure "points" per 100 possessions at a team level outside of ORTG.
So, there is an Impact Rating, prior to the Adj Impact Rating (AIR)? Is Jordan's AIR higher than anyone else's?
There is an expected impact rating that isnt adjusted for teammate strength, I didn't post those numbers though. That portion is easy enough to calculate. As for Jordan, I've only done 2013, and I don't think they have on-off data for Jordan lol
If there are any questions pertaining to how the stat was constructed or why a certain player got the rating he did, I would be glad to answer them.
I don't understand what it is supposed to show. I read the description and I looked at the results but it doesn't seem to tell me anything.
Why does Beverley have a better impact than Harden? I don't think there would be as much of a dropoff if Beverley wasn't playing as opposed to if it was Harden. Not sure if this isn't flawed (just like PER). Also, Casspi 4th most impactful player last year? WTF?
It shows how much of a positive (or negative) impact they make on a team's Net rating per 100 possessions. A player's role plays a huge part in the rating as you can tell from Bev's rating. There is some trouble with Collinearity looking at Chalmers' rating, but considering the other Heat players ratings look good, I don't think it is too big of a flaw in the metric. Im hoping that once I have a few more seasons of data I can group players by "role", and then do a separate regression for each "role" that should give more accurate results.
Its definitely not perfect, but I think its flaws are less substantial than PER's. Bev is higher than Harden because they have a different role, and the team played better with Bev on the court than Harden (+7.2 for Harden, +8.4 for Bev). This stat definitely requires someone knows a team's personal at least vaguely before deciphering, or you may come to the conclusion that Bev > Harden, which is definitely not what this stat suggests.
I understand what you're trying to do. You're trying to look at total Absolute_value[generated + subtracted] points-per-100possesions on an individual matchup (instead of team impact) basis. Individual instead of team impact is fine (best to have both) but I'm sure some would argue it doesn't have as much substance when looking at set offenses (Spurs) or team oriented defenses (Rockets). You use TOV and REB because they have the largest direct correlations to possessions (incorporates steals, mishandling the ball, etc.) and TS% to calculate how those possessions translate to points. Then you do a net calculation of offensive and defensive impact over 100 possessions (approx length of NBA game). I think the metric has value and probably done in some form in most league offices. I'd be suspect of FT/FGA and how you've gone about weighting REB, TOV vs. TS% as there are redundancies in their statistical value. Maybe people get confused because of the name Adjusted Impact Rating; I know I prefer more direct labels (maybe something like Net 100PPP Matchup Impact)
Ya i chose those variables because there are only so many possible results to a possession. Either a shot/freethrow is taken and made (TS%), the ball is rebounded (REB%) or there is a turnover (TO%). I used a linear regression to determine the weights of the variables so I didn't weight it myself. The only number that I chose myself was the .6 multiplier for the teammate strength adjustment, and that was through trial and error of just seeing what weight gave the best results. Good point about the name, I wanted to chose something that was easy to remember and abbreviated easily, but I wouldn't be against changing it to clear up any confusion.
net possession value is definitely a better standard for the overall value a player has while on the court. Better than PER as you stated (though Hollinger does adjust weight for league averages) and better than adjusted +/-. it would be awesome to see further study on team impact basis (how a player affects teamwide %), on/off, value vs. $, and something like impact above replacement.
and see I debated whether I should do impact above/below replacement or impact above/below league average. League average is obviously a lot easier to calculate (I don't know how to determine "replacement level"), so thats the way I went with the metric.
Makes sense what you're trying to do, but it seems extremely "noisy" in regards to role players. Jae Crowder kicked Dirk's ass in their respective roles. Patrick Patterson has a higher impact than Lowry and Derozan combined. I totally get your Beverley explanation for why that is happening, but I don't think there is any player in the league who could post those Patterson/Crowder/Babbitt numbers while playing starter's minutes. I also think this is certainly a nice PER alternative, but I'm also on record as pretty open in my disdain for PER. I also appreciate the attempt to make a plus-minus style stat a little more stat-based compared to margin-based -- but it does still run into the same flaws as most plus-minus based stats, namely: it is much easier to win small-minute battles against bench players than constantly win battles against the league's best as a starter, and there's not consistent correlation between succeeding in a role and succeeding as a big-minutes key contributor. In that regard, a strength-of-opponent modifier might add value, although I can only imagine how hard that would be to capture.
Thats exactly why at the end I added that I would welcome suggestions on how to add a modifier like that. Maybe adding a bonus for minutes per game over a certain point, and if they don't hit that amount of minutes per game it starts reducing their rating. 24 would be half the game so that might be the right amount to have as the "even", where players ratings wouldn't be impacted by the modifier.
well I figure if a player plays over half the game, his minutes are probably against starters for the most part, so the modifier I described above may work. For every minute above/below 24, they would receive a boost/reduction to their rating respectively. Figuring out how to weight it is another problem.
