Adjusted stats - how valuable?
View Single Post
11-02-2012, 06:07 AM
Join Date: Aug 2009
Location: Ho Chi Minh City
Originally Posted by
Czech Your Math
It's generally easier to destroy than create.
Many claim any perceived flaw means that adjusted stats are incompletely invalidated. They claim one of Mona Lisa's hairs is out of place, and that proves to them that the painting should be burned and we should go back to looking at stick figures. One of the stripes on the airplane looks crooked, so it's back to horses and buggies.
I, and others, have certainly made the effort to explain the foundation (value proportional to scoring context) and methodology (formulas) for adjusted stats. Specifically, while I prefaced my explanation of comparing players across the same range of seasons with the cautionary "this is difficult to explain" (and even more difficult to grasp if you do not fully understand the foundation and reasoning for adjusted stats), I still did my best to explain it by use of a simplified example.
I agree with your bolded statement. One would expect the two players to have roughly equal production over the combined two seasons, yet they wouldn't. The reason is that the scoring context changed substantially.
The player who scored more absolute goals will get more credit by people who don't understand that his goals didn't have more value than the other player's goals. Raw data's goal is to record the data as directly and simply as possible. There is no other goal or refinement present in the raw data.
There are various problems with comparing a player's production to a vary small subset of his peers (say the top 10 finishers or similar group):
- as has been pointed out ITT, comparative scoring between tiers can change and the reasons for that change should determine whether/how to further adjust for that fact (e.g., the top 10 players' adjusted scoring may increase simply due to being of comparative higher quality than before)
- the data for a very small subset is much more likely to vary substantially due to random factors, or for reasons that are more difficult to assess and properly adjust for
- the link between the adjusted data and value is broken
Simple adjusted data is built on the premise that goals win games, and that the value of a player's goal/point production is fixed in proportion to the average gpg in the season in which he was playing. Any deviation that does not explicitly and directly factor in the scoring context will distort the direct link to value which has been established. IMO simple adjusted stats should not be excluded in the quest for "new & improved" adjusted stats.
There are so many factors to consider for "new & improved" adjusted stats, that I'm not sure if/when there will be substantial agreement as to the proper method to obtain such. There was a thread in HoH about "Why would Gretzky still dominate today?" Well over 500 posts by many of the more knowledgeable posters and wide range of opinions on just how dominant Gretzky would be in today's NHL. That's just one player hypothetically placed in one different era. What about every player in any era? Coming to some sort of consensus on that will be incredibly difficult, but that doesn't mean it isn't worth attempting.
Pop culture certainly seems to maintain that its easier to destroy than create. Tell that to Faraday. The scientific method seems to be predicated in that belief to some degree or interpretation. Of course AS isn't quantum mechanics. That's a valid point here right? Its not quantum mechanics.
So why defend it as if it is? (Not to say you are. You present as quite open minded.
IMHO and experience when you have to do so much work to defend your idea then maybe its time to re-evaluate. I noticed that no-one gave a serious response to my point that ASs was likely, originally intended to predict what Howe would score today or in a Gretzky season. I remember doing similar stuff when I was 10 and reading about the history of hockey for the first time. I tried simple math, then I tried all kinds of new variables to try to fix some bizarre results. Then I resorted to reason. Later I discovered that when something works, it works immediately and always.
This is not about giving up coffee because a single study says it is bad for you. Its not about giving up on spaceflight because of a fire killing all the astronauts. Its not about refusing to sail the ocean because some believe there are sea monsters or that we'll fall off the edge of the world.
Its not that dramatic.
I appreciate yours and others(?) efforts to give reasons for your POV. Links (in the post) to specific studies referred to (in the post) would be appreciated. I think you might appreciate the results as well. Don't assume that I or any other poster knows wtf you're talking about. Be inclusive please and I'll try to do the same.
I kept looking at my statement that you bolded and thinking I should re-phrase that. It was not accurate. Instead I tried to explain in more detail where we differ. Its actually why I truncated your thread to the example.
Adjusted stats, raw stats, predictions, productivity expressions are not all equal. They can't be. They are each expressing a different aspect of the players results or productivity. Raw data has a special place in the hearts and souls of everyone. Ninety-two goals in a season is more than any other player has achieved. It is a benchmark. Nobody gives a rodent's hind end if Howe's production compared to peers was the same or better. They just see number 92 (imagine if he had scored 99? Was he trying to?). There is no number derived from any equation, no matter how convincing that will replace the number 92. Raw data is God.
Gretzky actually scored 92 goals in 80 games. Nobody has gotten very close. There is no possibility that any function, opinion or act of God will change that. Only a player scoring 93 will do that. And even then the circumstances he/she did it will (barring drug use) not matter.
My point is that AS are not nor ever will be raw data. So what do adjusted stats say? IMHO they were an attempt to compare raw data over eras and seasons but it didn't work and it just wasn't accepted by those not in the loop of those arcane calculations and some who were.
I'm pretty sure I posted in that Gretzky thread you referrenced. IMHO Gretzky scores more assists than the next best gets points and then 55-60 goals. Because that's what he usually, classically did in his career. Unless you're thinking of other variables such as transporting him instaneously, But then you're not taking into account that Gretzky had an effect on the game and so would be playing against himself so to speak.
You can't just remove a guy and his impact from the game by subtracting his stats from the formula. If he played then somebody, everybody and every point in between learned from him. You must see the league as if the guy never played, never impacted it and then guess what he'd do today.
I think the only way to do that is pretend that he never played and look at what he did when he played then induct what he would do if he played today. IOW I wouldn't use AS at all to predict. I would use reason.
What do adjusted stats do anyways if not attempt to predict? I'm still not clear on this.
Last edited by Dalton: 11-02-2012 at
View Public Profile
Find More Posts by Dalton