EUOC Big Weekend Ranking points....

[graeme]

I'll be seeing you at the meadows tomorrow then Mrs?

You're right, the old men were overrated in setting up the system. They're dropping slowly but beating up on them is the way to more ranking points. (edit: greywolf points out that winning urban races is better than winning forest races, on account of the reduced skew to long times. W21s in the lower half of the results dont seem to have shared the pointsfest)But as for weighting big events, take a look at the World Rankings and try convincing yourself that Doug had a better run finishing next to last in WOC than he did blowing the field away at BOC. And if you still believe in weighting, I have this bridge I'd like to sell you

Graeme, M??, deselected to M21, highest ranking score: Edinburgh Street race.

[greywolf]

the senior women's rankings are all fairly consistent for the big events, with the winner getting about 1250 and the rest spread as you might expect, but running a long urban race can score you massive points. without meaning to point the finger, i blame the vet men! they just seem to be disproportionatly bad at road running....

or maybe the top women, especially yourself, are very quick and can take advantage when the navigation gets easy - as someone with very creaky joints I'd have to say your lossie time last summer would be rather more challenging than the next day's pootle round Culbin

Given how close the top 9 women are in the rankings list, any amendments to the calculations would inevitably cause some shuffling of positions - and maybe if Level A events were weighted Hollie or Tessa would be on top - but the major effect would probably be to drop Sarah Rollins several places - just because she went to less Level A events than everyone else...

[AAH]

I think it works like this:

Draw a diagonal line from the top left of a bit of paper to the bottom right. The y-axis is 'points', the x-axis is time and the perfect straight line is people's theoretical time in an event where everyone performs exactly as expected. You can read off everyone's points for this theoretical event from the y-axis and you could also predict their time from their points by using the straight line fit the other way round.

Now imagine that was a very easy event that didn't really test anything other than road running and the points have been calibrated for that kind of event.

At a second event, there is some proportionate scaling of times due to length etc. but people are also separated more by a skill that wasn't really tested by the first race, e.g. Maybe it is quite technical. You expect the people at the top to be better at the more testing skill and people to be slowed proportionately more at the bottom end. So at this event if everyone ran exactly as predicted by their ranking the bottom right of the straight theoretical performance line would actually curve off to the right hand side of the bit of paper. If you now plot a line of best fit through this curve the straight line goes below the curve at the left and right of the sheet of paper (fastest and slowest runners). So they score fewer points that you might expect if the points allocation was calibrated for the first type of event.

At a third event, people are separated by all sorts of skills including terrain running, technical ability etc. It's long and some people get tired and start walking who wouldn't usually. This time there is a really dramatic curvature - some people who might be expected to take 150% of the winner's time at the first event might take 200% or more. Again, if you plot a straight line through this to allocate points, the fastest and slowest get fewer points than you would expect...

So the more skills that are tested, the more the scores of the fastest and slowest participants are decreased as a result of the increased curvature of the line of ranking points versus expected time.

In practice, the BOF ranking list is set up for something like the second event. That means that for events that test fewer skills (e.g. Street races) points at the top and bottom are inflated. For very difficult events, points at the top and bottom are reduced.

Thus, the top scores at major events last year were, for the women:

BOC Long: 1216
Bigland: 1219
JK Middle: 1222 (Tyrella dunes...)
JK Long: 1238
Bigland: 1246
British Middle Distance: 1247
JK Sprint: 1256
British Sprints: 1280

And for the men...

Bigland: 1313
JK Middle: 1315 (Tyrella dunes...)
JK Long: 1318
Holme Fell: 1322
BOC Long: 1320
British Middle Distance: 1332
British Sprints: 1341
JK Sprint: 1357

So I think that the effect housewife refers to is more to do with fitting straight lines to curvy data than vet blokes road running skills...

Possible fixes without major overhaul: where there is enough data, only using the top 20. Where there is enough data trying to plot a curve. Weighting the scores used for ranking heavily favouring those at the top end so at least their scores are right.

[greywolf]

AAH : there is a linear realtionship but it's centred on the mean and calibrated by standard deviation, so it's the relative distribution of times that matters. The total aggregate points scored at any race is a function of the existing points of ranked runners - but how this total number of points is distributed is dependent on the spread of finishing times.

Say you have 23 eligible ranked runners finishing a race. The bottom 3 are excluded as outliers. If other 20 runners finish in 60, 61, 62 minutes and then 1 minute intervals all the way to 79, then every single one of them would score exactly the same points as if they'd finished in 60, 62, 64 minutes all the way to 98. Sounds rather counter-intuitive - but each athlete is the same fraction of Standard deviations away from the mean (and the value of mean points and SD points is set by their pre-existing scores)

On the other hand, if the same runners finish 60, 62, 64, 65, 66, 67, 68. 69, 69, 69, 70, 70, 70, 71, 72, 73, 74, 75, 77, 79, the mean is the same as the first example but the SD is smaller and thus the winner would score more points than in the previous two examples, because their time would now be >2 SDs from the mean, instead of c. 1.65

I think that in technically simple races the field doesn't get spread by major mistakes, and even weaker navigators can run closer to their maximum pace, and so the majority of finishers have times clustered around the mean, certainly looks as if that was the case on both the A & B courses on Saturday

[NeilC]

Last year I made the point to a leading member of the ranking working party that the reason given for excluding night events (an atypical results distribution) probably also applies to urban events (albeit atypical in a different way) and therefore either both event types should be excluded or both included. I should point out that I was asking for a valid justification for excluding night events, not looking to exclude urban events.

[andypat]

Absolutely. Night events shouldnt be excluded just because some people arent good at it or dont like them!

[AAH]

Greywolf: the formula using mean times and the standard deviation of points around the mean works fine provided that everyone’s expected times scale in a linear fashion for an individual event. But I think it all breaks down if they do not.

Suppose runners 1 to 20 have existing average ranking scores of 1300, 1290, 1280, 1270 etc., you would expect them to have equally spaced times at an event (ignoring random variation around expected performance). I agree it doesn’t matter if it’s 60, 61, 62, 63 mins or 90, 91, 92, 93… or 60, 63, 66, 69, 72… or 15’00, 15’20, 15’40, 16’00… etc. If their times are equally spaced, they all get allocated scores 1300, 1290, 1280, 1270 etc. Any linear transformation changes nothing.

Now suppose the distribution is not linear at some events. Some additional factor (e.g. slow runners getting tired and slowing down by a larger percentage than fast runners, or poorer navigators making significantly more mistakes, or people who have made mistakes stopping being bothered) means that instead of runner n taking e.g. 59 + n minutes (60, 61, 62, 63…) they take 59 + n + n squared/ 10 minutes, i.e. 60.1, 61.4, 62.9, 64.6 etc.

When each runner took 59 + n minutes you get a mean of 69.5 mins and a SD of 5.916 mins. Add in the square term and you get a mean of 83.85 mins and a SD of 18.590 mins. The mean of the runner’s existing ranking points is 1205 and the SD of their scores is 59.161 in both circumstances. Working out the ranking points, that means that the first runner gets 1300 as you would expect when each runner takes 59 + n minutes (or indeed a + bn minutes) but only 1280.6 in the second circumstance.

I’ve put together a jpeg to show this but can’t seem to upload it at the moment… I’ll try again later.

The formula using standard deviations does struggle as the data being fitted is very non-Gaussian. It gets away with it usually because the distribution of times at an event should be skewed in roughly the same way as the distribution of score of the people who are present. But whatever straight line fit is employed, there simply isn’t a way to fit a straight line to a curve. If something happens to cause the lower placed runners times to drop off disproportionately, points at the top end suffer. If some factor which usually holds the lower placed runners back disproportionately disappears, the top placed runners get extra points.

[Becks]

Tired people from the Devil's Burdens going round slower than usual? Tullster, for example, is well down on the Men's.

[graeme]

the data being fitted is very non-Gaussian.

If it were Gaussian, about 2.2% of runs would score more than two standard deviations above "average" (i.e. 1400+). There's around of 50000 runs in the statistics, so there should be over 1000 such high scores.

The actual number is, er, zero.

[The Loofa]

Actually I think all runners on the B course should get even more points. Look at the runner who took 44.59. On the EUOC website this is Richard Oxlade; on BOF results its his wife, Anne Hoy. Given EddieH was one place lower its safe to assume this was Richard running and equally safe to assume his average is much higher than Anne's. So if the BOF results get corrected I think that means all runners on that course get a higher score - but probably not by very much.

[The Loofa]

I also want to add my support to NeilC's argument about night events. There seem to be quite enough other anomalous things floating about and they all seem to get evened out in the long run. So I cannot see any justification in excluding them.

Problem is what to do about it. There seems to be no body (committee/whatever) in existence that could address the issue. Unless someone knows better. Could be a task for SuperScott.

[SIman]

Richard is now correctly in the BOF results (NUL and SOUL) will be changed later.

Wouldn't it be nice if the BOF system had some form of validation built in - or even allocated a BOF number if the name, club and age were right

[greywolf]

Actually I think all runners on the B course should get even more points. Look at the runner who took 44.59. On the EUOC website this is Richard Oxlade; on BOF results its his wife, Anne Hoy. Given EddieH was one place lower its safe to assume this was Richard running and equally safe to assume his average is much higher than Anne's. So if the BOF results get corrected I think that means all runners on that course get a higher score - but probably not by very much.

Not that simple... Correcting the error (if that's what it is) will increase the mean points awarded - crudely if Richard's average points score is 400 higher than Anne, then that's an extra 400 points to be divvied up; but it will also affect (decrease) the SD of existing runners' scores, which affects points distribution (each SD from the mean will be worth a few points less), so those at the front will probably end up with less points.

[edit: just checked and glad it did. that's still a lot of points, mind]

[distracted]

Cat among pigeons time...

Using a version of Greywolf's example in a hypothetical situation - two different courses with the same field competing are won by the same person, both in 60 mins. Clearly the mean/sd ranking of both fields is the same.

However, Race 1 is in tougher/more technical terrain and has mean time 90 mins and SD 15 mins, the stats for Race 2 in easier terrain are 70 mins/5 mins. So the winner gets the same number of points for both races

I'd argue that the winning performance in Race 1 is more impressive than that in Race 2. Surely that race should gain more credit when it comes to ranking points? But not with the current system...

[awk]

I'd argue that the winning performance in Race 1 is more impressive than that in Race 2. Surely that race should gain more credit when it comes to ranking points? But not with the current system...

And I'd ask why? Even if one believes in weighting of races, which I am no longer sure I do, just because a race is a bit longer and in harder terrain doesn't make it any more important.