Ranking List

[DeerTick]

Wow, I'm a better orienteer than I thought. Was thinking of retiring, but what a boost!

Yesterday I remembered 1100+ scores the way I remember long, thick hair, desert boots, Jefferson Airplane and five minute miling.

So 1100 scores are here again. The rest I'll still remember. OK, I haven't risen in the ranking tables, but it's a warm feeling - I'm not ageing as rapidly as I thought.

AP

[MIE]

At first glance it seems like positions are entirely unchanged.

Nope !

Up 738 points and up 22 places

[DeerTick]

Yet still the results are complete sh8t
What's the point of having a combined ranking list when it's so inaccurate. You think with the collective wisdom around, something better could be done.

Can you give some evidence for either of your assertions?

AP

[awk]

No change to the algorithm. The main change is that the list will be re-based regularly to cure the downward drift of points problem......Should be retrospective as well ... I hope

Pretty much everything should go up. There was/is a flaw in the algorithm which causes a steady deflation of scores over time. So recent scores will go up more than old ones.

The following are all the dates where I scored between 1100 and 1110 inclusive:

Date - old score - new score - difference

17.04.2010 - 1103 - 1178 - 75
30.01.2011 - 1109 - 1183 - 74
09.04.2011 - 1103 - 1178 - 75
10.07.2011 - 1103 - 1178 - 75
29.10.2011 - 1108 - 1182 - 74
21.01.2012 - 1102 - 1178 - 76
01.04.2012 - 1105 - 1180 - 75
22.04.2012 - 1106 - 1181 - 75
12.05.2012 - 1105 - 1179 - 74
24.11.2012 - 1104 - 1179 - 75
26.01.2013 - 1108 - 1182 - 74
10.02.2013 - 1107 - 1182 - 75
16.03.2013 - 1102 - 1178 - 76
30.03.2013 - 1107 - 1182 - 75

So it looks as if it hasn't dealt with the deflation issue - and that was the biggie.

[graeme]

Well that's very interesting (to those of us who find such things interesting). It looks like they've just rescaled everything (points and standard deviation) rather than doing "a complete re-run of the results data".

Have you got your whole set of old and new scores?

[awk]

Have you got your whole set of old and new scores?

I ran a print off of all my scores from Ollie's website a couple of days ago in anticipation, and then recovered the new scores from the BOF results pages. I looked at these as most of my scores lately have been running at round the 1100 mark. It'll be straightforward to put a set together once Ollie's site shows the new figures.

[DeerTick]

Well that's very interesting (to those of us who find such things interesting). It looks like they've just rescaled everything (points and standard deviation) rather than doing "a complete re-run of the results data".

Yup, that was my suspicion.

AP

[King Penguin]

Ollie's stats have not updated this morning for the old events (before approx. mid-March), which makes me suspect he might have to do something to force a full reload.

[Clive Coles]

To go back to the original post

On Wednesday 3 April 2013 the ranking system will be updated. ..................

Hopefully the process of updating the system will be smooth but we have learnt from past experience that this may not be the case.

The ranking leader table is recalculated every Thursday morning consequently the leader table as presented on Thursday 4 April will reflect the changes.

Looks to me this update has been planned as a two stage process ~ with a "hopefully" and "this may not be the case" inserted in the explanations to set expectations.

Could well be therefore by the end of today things will have changed ~ once the full update process has been completed

[charles2]

Don't know why they've done this with the rankings, but the relationship between old and new scores is linear and it is roughly

new = 0.8 * old + 298

old scores of zero are now being given as 298.

The percentage increase of an old 660 score is 25%
The percentage increase of an old 1000 score is roughly 10%

[King Penguin]

Rather concerning if the previous points of 0 were due to mis-punch, retiring or non-comp.

[charles2]

If you mis-punch, retiring or non-comp then your score for that event is blank so that is unaffected (it is still blank), if you finish an event your points were max(calculated pts, 0), i.e. couldn't score less than zero.
But going forward you still cannot get less than 0, there will now be competitors scoring less than 298.

[funnyrunner]

I can try and give some evidence, unfortunately It's going to be a bit specific - Apologies to C Hallet.
Basically there is a certain amount of self fulfilling prophecy in the rankings.
For example many of the best "old man" scores come form local events. There then seems to be a lack of cross pollinisation between the different groupings so you can replace old man by Elite girl or Junior.

For example w.r.t Clive Hallet - there is no way he is the 34th best orienteer in Britain. Although I recognise he is a bit of a legend.
His top score comes from a BOK level C event where he beat..... nonone. Basically he just smashed everyone there, and got his highest score as he already had a high score. Interestingly the 2nd place getter also got his highest score there. Therefore it's self sustaining.
As soon as clive runs with in an event with decent M21s he doesn't score that highly and the scores get disregarded. The M21 doesn't get the "clive effect" if another M21 turns up as the overall scores then get pumlled. It just seems too sensistive at small events.

This example is quite specific, but generally leads to old men/elite women/junior boys being overanked. Whilst it not perfectly fixable due to low cross pollination, I think it surely can be mitigated. There must be some strength of field indicator, such that you would expect the top scores to be produced in races with a sizeable number of decent athletes. Which should remove statistical anomalies that plague the results.

Additionally there's the recognised problem if someone decent turns up and walks round the course, but again, surely one outlier shouldn't influence everyone's results.

I can't provide any evidence on the collective wisdom of nopesport readers.

[SJC]

His top score comes from a BOK level C event where he beat ......

This was always going to be a problem once all the level C events were included in the rankings. Clive has probably run on that area every year for the last twenty years, knows where all the control sites are, and as a consequence is going to get pretty close to a perfect run. The majority of the competitors perform at their normal level and so Clive gets a very high ranking score.

[King Penguin]

How about using each individual's 2nd to 7th best scores rather than 1st - 6th ?
Where someone has just one unusually high score that anomaly would be removed.
Granted, someone with 2 unusually high scores still benefits.