Event Structure Review: Rankings List

[Adventure Racer]

Some clubs offer a 3-in-1 series of 3 events on the same day, and a few years back SEOA offered 4 events around the M25 on the same day. You choose how many to go to, and in which order.

Surely those sort of events would count as local events, and therefore be non-ranking (or am I wrong in thinking that only standard/district events and above would count for rankings?)

Unsurprisingly I'm in favour of night events counting for ranking - I'd be even more short of counting events otherwise! Whilst I take the point about distortions due to big mistakes, it is also the case that night-o requires slightly different technique and therefore practice, and given I possibly do more (local) night events than day events I feel I have earned any advantage I might have. Having a separate night ranking would get a little silly with sparsity of results.

[mikecumpstey]

Post by graeme on Thu Oct 11, 2007 8:36 am

Of course, we did produce an algorithm which we believe will be stable, details of which are at

http://www.ph.ed.ac.uk/~gja/Competition ... xK_rev.doc

This document specifies the proposed calculation. It is very similar to the current scheme except that the calculation is based on median/quartile statistics rather than mean/standarddeviation used in the current scheme. As it stands the specification has several inconsistencies which I take to be errors but it easy enough to get the general meaning.
I would like to be told the justification for moving to median statistics rather than mean, which is used in the current system. It seems like a good move - a way to overcome many of the problems the current system has. If you want to see how it pans out I have set up a full single class ranking list using the new statistics (URL = http://ccgi.orienteering.plus.com/ranki ... wlist.html). It uses data from ~600 events since January 2000, so the problem of kickstarting is insignificant. In any case the suggested method of kickstarting makes no sense since scores are linear with time so should not be multiplied by a speed factor. No event weighting has been used.

[graeme]

Since you ask...

Median/quartile statistics should be better than mean/variance because the time distribution is very skew towards long times. Under mean/variance statistics very slow times have a wholly disproportionate effect. The present list tries to address this with various complicated fudges, all of which (should) go away with median/quartile stats.

I'm not aware of errors, but this doesn't mean there aren't any, can you pm me details?

Startup system isn't pretty, but the average score is scaled to the mean speed in each age group - so the scaling factor for mean score is about the only thing that is right about it! Note that there is a "proper" ranking system hidden in there for calculating the points, as well as a ranking skewed by discards (and, in principle, weightings) which provides a better "virtual competition".

Graeme (still disinterested)