RealGM

I have created a new top-down metric which I call Adjusted Impact Rating. A description of the stat, as well as the ratings for the 2013-14 season can be found at adjustedimpactrating.weebly.com
Feel free to ask any questions about why the formula is set up how it is or if you are curious why a certain player received a certain rating.

So no questions about ratings or how the formula was created? I know i'm not THAT good at explaining this stat haha.

Okay, so I've taken a look at it. I think I understand it. Basically you're trying to create a PER-esque model by using on/off-based regression instead of actual individual stats. Is this right?

If so, I'm impressed with the skill involved in creating it, but I have to ask: What information are you hoping to get out of it?

I can see a perspective that would say that a metric like this is superior to PER and its ilk based on a focus of impact rather than something that credits guys for their hoarding of stats. If done right, with ample sample size, this would indeed be a better metric.

However, I see no reason to think it would be a better metric than actual RAPM stats. Seems like you're just adding distractors from the true value given that if your metric has ample sample size to be trusted, then so does RAPM.

Thoughts?

Last thing I'll say is that I'd really like to see more studies done using regression on particular stats individually, like you're clearly capable of doing. I don't see how work like this replaces RAPM, but it could certainly produce, say, a more informative list of defensive rebounding and turnovers than the raw data can give.

@Doctor MJ
The goal of Adjusted Impact Rating is to show how much impact a player makes in "on Net Rating". I think I still need to add some adjustment that penalizes/rewards players for playing under/over some pre-determined minutes per game respectively. Once I figure out what that threshold should be and how to weight the over/under on it, it will be really simple to apply it. This will bring players like Crowder, Harris, and Babbitt closer to the mean while rewarding the players who play huge minutes. Maybe even an escalating reward/penalty would be appropriate.

Also I plan on doing a boxscore stat soon, but I'll also incorporate sportsvu to get a better defensive view than most current stats/

houstonredox93 wrote:.

OP,

Can I ask why you adjust each individual component by comparing with league average as opposed to team average? Because what it looks like to me right now is that this stat is only really accurate in rating players on the team relative each other, and not for comparing across teams.

For example, every single player on the 76ers is listed as a negative, and this is likely because the 76ers were so far below league average in every stat. But the implication is that all of these players hurt the 76ers when they were on the floor, which logically cannot possibly be true. It's just that they weren't good enough to improve them to league average. Ditto for the Pistons, I mean were Drummond's minutes really killing the Pistons to this extent?

Unless you intended for this stat to be biased toward good teams, in which case this makes some sense and you can ignore this.

Doctor MJ wrote:Last thing I'll say is that I'd really like to see more studies done using regression on particular stats individually, like you're clearly capable of doing. I don't see how work like this replaces RAPM, but it could certainly produce, say, a more informative list of defensive rebounding and turnovers than the raw data can give.

And I was really hoping to see you respond to this paragraph, OP, because if I read MJ correctly, what he's asking is really really tantalizing. And your response tells me you might not have picked up on that.

Basically, he's asking if you've thought about creating a metric like this, but with a narrower focus than this all-in-one thing. So for example, you could separate each component of AIR and do, say, "Rebound +/-", which would be a stat solely to measuring a guy's impact on team TRB% while he's on/off the floor. Same thing with TOV%, TS%, etc. Essentially, split this up into its individual components and see what happens.

@Dr Spaceman
I compared to league average instead of team average because if I compared to team average, teams like the 76ers would have close to half their roster as a positive. Do you really think that the 76ers had a bunch of players that were above league average? And they did have one player rate out positively, Brandon Davies was a +1.656. I'm about to add a new part to the metric that will reward/penalize players for playing over/under 24 minutes per game that will probably make the ratings look closer to what you were apparently hoping for. The purpose of this addition is to reward players for playing against better competition, and im following the logic of "The more minutes you play, the more likely you are playing better players". Its not perfect logic by any means, but I think it will do a lot more good than harm to the overall ratings. One example I can think of off the top of my head that wouldn't go by this is someone like Joey Dorsey this season lately. He plays only 12 minutes or so a game, but they are against starters. In this case the adjustment would make his rating less accurate, but I feel confident that the opposite will be true in most cases.
And the stat is bound to be biased towards good teams if it is doing its job since good teams usually have more good players than bad teams do. I'm not really sure why you expected the 76ers to have a lot of players rated well in a stat that rewards good play.

As for the narrower focus, I've thought about the rebounding version, but that is the only one I think I could do with the stats currently available to me. If I plan to base it off a regression similar to this stat, I would ideally want more than just a couple input variables. That will probably be a couple more years down the line unless a surplus of stats are released to the public.

Made the new adjustment and updated the site.