Originally Posted by
fantasyhockeygeek
I thought I heard my ears burning!
Thanks for the insights!
Nope, for a couple reasons.
The first is that you're making an assertion that it doesn't make sense to rank above-average D over average F. The value on FHG is actually "value over replacement", ie value as compared to the best option available on waivers. This "replacement player" varies greatly based on league size.
This was poorly stated on my part, partially because I was trying to recreate to some extent the way this objection has been raised by DobberHockey posters in discussions prior to their learning more about such player evaluations. In particular, my use of "average" was far from clear. I know that the valuation represents value over replacement, but I think many people have wondered whether it is true that if defenseman A is 30 "points" better than replacement (specifically the strongest non-rosterable player) while forward B is 5 "points" better, if the defenseman's value to your team is really greater than the forward. Of course the replacement varies with league size and roster configuration, but those factors are already included in the value calculation
I'm not sure that basing things of "ranks" is fair -- by doing it that way, you ignore gaps in valuation ie you're assuming that the gap between player #1 and player #2 is the same as the gap between player #2 and player #3, aren't you? That, to me, is a huge misrepresentation of player value.
Excellent point. I think this shouldn't be a big concern if the shape of the drop-off is similar between forwards and defensemen, because then the large gap would place the players in the correct order. For example, if there is a big gap between D #3 and D #2, that gap is needed to push D#2 up into a rank between, say, F #7 and F #8. Of course as you point out the assumption that the drop offs are similar is a bad one, in which case I'm not sure if the argument above holds.
Wait, now that I read your answer more carefully, I see that your actual point was simpler. I think in my mind I justified using ranks instead of actual values as representative of how one would use the valuations in a draft, i.e. simply draft a higher valued player over a lower rated one. But of course, as I explained to the OP in my previous post, that's not how these valuations should be used anyway, so I'll just shut up now. This is apparently one of the dangers of trying to practice Python programming with the flu.
If you're looking at a population of 60F and 40D, you're looking at roughly the top-2 forwards on each NHL team and the top 1.33 D, right? Usually, D are pushed up in value by having more of them owned, because below the top 60ish you see a big drop-off in production.
I see your point here, but for any specific league set-up, shouldn't the valuation still be normalized across positions? In other words, doesn't the value calculation already include the number of forwards and defensemen rostered in the league, and thus a player with a higher value is more valuable to a team than a player with a lower value, regardless of the fact that one of them is a forward while the other plays defense? As Campkin discusses, the different characteristics of the drop-off curve mean that different positions' relative values will change when comparing different league formats, which I think is part of what you're saying here. But shouldn't we not have to worry about that within one particular league?
I guess I'm having a hard time with your drawing a general conclusion like this based on one specific setup, when the correct answer is that the value of defense is dependent on the league size and scoring categories.
Another huge problem with my post. The ambiguity in my use of "average" makes the question sound less connected to the actual meaning of the evaluation than I meant it to be. I meant something along the lines of average valuation under the evaluation system, not average in production. To top it off I then said it all backward as I mused if the results, if true (no evidence of this anyway), would cast doubt on the assertion when I should have asked if it lent any support to it. And that doesn't even take into account the rank vs. value problem you point out. Further, I never meant to imply that such a result (again, if it actually happened) would be in any way be able to be generalized to any other league format, especially, but not limited to, multi-cat leagues. And even just for points-only leagues, this is not very representative, as I think this is an example of pseudo-replication:if there were a trend of such results occuring with the same league format but for multiple years' data sets, then it would be a little more interesting. Among other reasons, looking at only one year's numbers is particularly vulnerable to the rank vs. value problem.