View Single Post
Old
11-01-2012, 11:18 AM
  #143
Iain Fyffe
Hockey fact-checker
 
Iain Fyffe's Avatar
 
Join Date: Feb 2009
Location: Fredericton, NB
Country: Canada
Posts: 2,659
vCash: 500
Quote:
Originally Posted by Rhiessan71 View Post
That still doesn't change my point about averaging and the increasing inaccuracy produced the further away from the median you go.
That's the claim, yes. Looking forward to the evidence that it's more than simply an assertion.

Quote:
Originally Posted by Dalton View Post
It seems to take a lot more work to defend adjusted stats than to critique them.
It does if the criticisms are imagined things that take no effort to come up with (because they are assertions, rather than conclusions based on research, which take effort).

Quote:
Originally Posted by Dalton View Post
Scoring 49 goals in 70 games compared to 92 in 80. Obviously 92 goals is the greater achievement but if we want compare the players production compared to their peers in any season we see that 49 in 70 was a little bit better. At worst they are comparable.
Obviously 92 goals is the greater achievement? Aren't you missing a whole lot of context before being able to make a claim like that?

Quote:
Originally Posted by Dalton View Post
I would also remind that this is not a math question. We are talking about evaluating and predicting production.
...by using math. Evaluating, yes. Predicting, no. See below.

Quote:
Originally Posted by Dalton View Post
We have a study that shows us that outliers play a large role in determining averages. Hockey GS among left wingers were used in the HR study. It's like a fractal. Outliers influence the averages regardless of the sample size. Teams, divisions, seasons all the same. Using averages results in errors of several magnitudes higher than not using them.
Are you referring to the study you linked to? It doesn't say you shouldn't use averages. It says that outliers should not automatically be disregarded when calculating the average, and that human performance should not be assumed to be a normal distribution. Adjusted scoring does neither of these things.

Quote:
Originally Posted by Dalton View Post
In some seasons the sum of production of all the players is an outlier.
You don't know what outlier means, if you make this claim. We discussed this earlier in the thread. There is no strict mathematical definition of what value would constitute an outlier in a particular case, but if a study suggests an entire season's worth of data is an outlier, then said study should rightly be criticized for that.

Quote:
Originally Posted by Dalton View Post
The mere fact of using an average to calculate production should lower the outliers performance at the high end and increase everyone elses by definition.
And yet, it doesn't. So perhaps you're not as clear on what's happening as you think you are?

In a season with 5 GPG against an arbitrary base of 6 GPG, everyone's production will be increased (everyone with more than 0 goals, that is). Those below the season average are pulled toward the average, while everyone above the average is pushed away from it.

Quote:
Originally Posted by Dalton View Post
The 'average' worker's production is actually below the average production of all workers because of the effect of the outliers on the average.
This has nothing to do with outliers, it has to do with the shape of the curve. As I've said, in a power-law curve the "zero" results are not outliers; only very high results could be considered outliers (and they're not necessarily outliers), and due to the small number of such results they have a very small effect on the calculation of the average.

If I'm reading you correctly, you're saying the mode is less than the mean in a power-law curve? That's true, but obviously not a problem since we should be using a power-law curve, yes? It's something that shows up in the raw data as well.

Quote:
Originally Posted by Dalton View Post
Did I do that correctly?
It's a bit late to be asking if you understand the very basic concept of adjusted scoring, isn't it? If you're not sure, perhaps you should study it some more before trying to discuss the finer points.

Quote:
Originally Posted by Dalton View Post
I think people started this adjusted thing to actually predict what each player might score in each other's season's as a means of comparing them.
That's not a prediction. A prediction would be using the adjusted scoring number to guess what the player might do in a season that hasn't happened yet. Adjusted scoring is not intended to make predictions any more raw scoring is.

Adjusted scoring is a means of more fairly comparing two players' seasons when they were different scoring levels etc. No more than that.

Quote:
Originally Posted by Dalton View Post
Criticism has moved everyone away from this original line of thinking. Unfortunate because I believe it is the source of the debate. Adjusted scoring was meant to predict
Prove it, then. Adjusted scoring was first done, I believe, in Total Hockey. This is, of course, a historical encyclopaedia of the NHL. It seems pretty clear to me that the system was developed to allow a player from 1929-30 to be more fairly compared with one from 1989-90. In fact, we don't actually have to guess why they did it, because they tell us why they did it:

Quote:
Originally Posted by Total Hockey (2nd Ed), p. 613
In preparing Total Hockey we set out to create a reliable method of comparing players from disparate eras.
Emphasis added. A method for comparison, not prediction.

Quote:
Originally Posted by Dalton View Post
I don't believe adjusted stats have any role anymore now that it seems to be accepted by all that they don't do what they were originally intended to do- predict.
What are you saying now? That some people don't understand what they're intended to do, so now "all" have "accepted" that they're useless? You're digging yourself deeper here. This statement is flatly false.

Iain Fyffe is offline   Reply With Quote