Discussion: Rankings!

So instead of working on my thesis this weekend, I've built the USOF ranking formula into attackpoint. I think I got it right, but tell me if you experts see anything suspect. The scheme uses all the results posted here, without distinction by 'color', as long as there are enough people on the course.

click on 'rankings' in the left menu.

Very cool... People seem to be ranked right about where they should be... But I noted Ernst Linder way down at the bottom around 40 points. Definitely not Ernst... probably due to only having a few races and those races appear to have been against a relatively weak field - none of the regular high ranking runners to compare against probably yielded low course scoring values.

I presume the smaller type data are the courses that get dropped by virture of running more than the base limit of 4 events?

Ken, you may want to make sure you use the harmonic mean instead of the 40%le. The harmonic mean is much more stable. Yum, the harmonic mean.

Nice and cool! But irrelevant to real ranking since not everybody who competed entered data. Specially Canadians who try to avoid Attackpoint :) At least it is valuable tool to see how you performed relative to others in a race. Thank you Ken!

yes, the idea is obviously to rank the attackpoint community using the usof method, not predict your usof rank. but considering that the rank points are proportional to racing speed, you can expect relative scores to be totally relevant (e.g. 100 is still twice as fast as 50) with the only actual difference being a matter of normalization.

sorry about those 1-pt west-coast rankings, using the harmonic mean diminished that effect somewhat but introduced other wierd ones, so for the moment I'm still using the average of the 3 PSVs closest to the 40th percentile.

The Divergence is Near! Run for your life!
65. Adrian Zissos 28.3 67.94 , 42.12 , 2.65 , 2.0

There's also a very fine point, Kenny, in calculating PSV_i which is not covered in USOF 9 (a). The "previous ranking" mentioned there can mean one of two things: a raw average of one's scores from iteration i-1, or the average of the 4 + floor [(n-4)/2] scores from i-1. Bill argued a few years ago that it is the former that yields the cleanest convergence, and to the best of my knowledge, he still uses it, although the latter seems to conform to the spirit of 9 (a) and 11 best.