Quote:
Originally Posted by Sturminator
I see. Well, this is why it is important to be careful about checking our data and communicating exactly which methodology we have used.

Honestly, I threw my numbers together rather quickly. Based on the switch I made after season #3, it was too quickly.
Quote:
The 1992 season is a difficult one to evaluate. Gretzky is only a cunthair over the 10% standard as compared to Hull in 4th place, and Stevens is in the top3, which smells funny. I really don't like using Hull as the standard here because of Stevens. If Gretzky was such an outlier in this season, there's no way Kevin Stevens outscores him, even playing with hockeyjesus on his line.

I just remove Gretzky and Lemieux regardless.
So I looked at Stevens dropping to Hull, and it was more than a 10% gap.
Quote:
The problem here, as well as in 198889 and 197071, is that you end up with a distortion no matter who you use as a benchmark. Stevens as a benchmark likely results in a downward distortion of Vs2 numbers, but Hull as a benchmark likely results in an upward distortion.
My gut tells me we need a special standard for dealing with seasons in which there is greater than a 10% gap in the top5 after the 2/3 gap. In these seasons, we get into territory where there was a whole group of players who are treated as outliers, and that raises the question of whether they really were outliers, or whether the rest of the league just didn't score that well. I would propose counting the number of players in the "outlier tier" (so in 199192, there are three), and averaging the scoring of the topx players in that season as a benchmark, with x being the number of players in the outlier tier times two.
So, for example, in 199192, we would average the following:
1. Lemieux  131
2. Stevens  123
3. Gretzky  121
4. Hull  109
5. Messier  107
5. Robitaille  107
...to come up with a Vs2 benchmark number of 116.
So in Turgeon's case, we end up splitting the difference, which is probably the least distorting result.
My method: 95/123 = 77
Yours: 95/109 = 87
New proposed method: 82  which almost exactly splits the difference.

This stuff is hard, and we really need some kind of standardization. The Vs2 methodology is quite clearly superior to the ranking system, but it would be very helpful if everybody agreed on which numbers we should use. I will have to do the hard work of actually going through NHL results seasonbyseason as seventies has suggested in order to further the discussion. Ugh...

That's exactly why I use a combination of rankings and percentages. There is no perfect method.