22 replies to this topic

[hatless]

G’day everyone,

Given the hot weather lately, I thought I would do some analysis on the impact of the weather on the results for our favourite trail run.

Methodology:
Temperature data: I got the hourly temperature data for the Mt Boyce AWS on the days of the race. I took the average of the 8am to 2pm readings to get the “average” local temperature over the time that most people are out on the course. No data was available before the 1990 race, nor for the 1997 race.

Race time data: I took the median and the 25% and 75% quartile times of the number of people who started. That is, if 800 started, I took the 200th, 400th and 600th time.

Years covered: I looked at three sets.
1990-2008 covered all the data I had available (although I removed 1991 because it was a big outlier.)
1996-2008 was interesting because I suspect that some of the earlier years are not representative of the range of people who enter the race today.
2001-2008 was done because that is when the last change in course and time limit happened, although the temperature range since then has been pretty narrow so you don't get a lot out of it.

Results:
For each of those data ranges the average 25% quartile, median and 75% quartile race times were:
1990-2008 5:06:02, 5:41:58, 6:20:38
1996-2008 5:10:05, 5:45:47, 6:23:03
2001-2008 5:06:53, 5:42:25, 6:19:44

and the sensitivity to a one degree change in temperature is

1990-2008 0:02:14, 0:02:14, 0:02:28
1996-2008 0:02:28, 0:02:49, 0:03:37
2001-2008 0:02:04, 0:02:44, 0:03:42

In case you a wondering what all this means, if you are, say, about a 5:10:00 6FT runner, then (based on the 1996-2008 data set) you should expect your time to increase by about 0:02:28 for every additional 1 degree in temperature - so a day 2 degrees hotter than average should yield a 5:15:00 time. If you are a 6:20:00 runner, then on a 2 degree hotter day you should run about 6:27:00.

It also means that the difference between the hottest and the coldest days of the race (about 10 degrees difference) should yield times around 27mins different for the median runner.

Summary:
For every 1 degree change in temperature, runners should expect to add 2-4mins to their final time. So a 5 degree hotter day should add 10-20 mins. Just out of interest, last year was about 3 degrees cooler than the average, which gave most people a headstart of around 6-12 mins.

Cavets:
None of this analysis takes into account other factors such as humidity, rain, wind, effects of a muddy track, depth of Cox's river, trees down in Nellies Glen, changes in cut-off times, changes in the route, changes in the distribution in people running the event etc. However, looking at the attached diagrams, the temperature seems to be a pretty good indicator of the median time. With an R^2 of around 80%, Occam wouldn't be recommending anything more complicated.

If anyone wants to do their own analysis, I am more than happy to fwd the source data on.

Good luck with training over last few weeks,

h.

Attached Files

•  6FT_1990_2008.jpg   43.54K   73 downloads
•  6FT_1996_2008.jpg   41.72K   55 downloads
•  6FT_2001_2008.jpg   39.7K   39 downloads

Edited by hatless, 11 February 2009 - 02:44 PM.

[nando]

WOW!  That's some work you've put into that Hatless.

Are you able to publish what the average temps were for each year, so we can factor in this for our own times.

Thanks

[hatless]

nando, on Feb 11 2009, 03:55 PM, said:

WOW!  That's some work you've put into that Hatless.

Are you able to publish what the average temps were for each year, so we can factor in this for our own times.

Thanks

Here is a list of the dates, the "average" temp and the 1st quartile, median and 3rd quartile race times. (Excuse the lack of formatting - these tables always turn out ugly...)

8/03/2008         15.2 4:57:28 5:27:55 6:00:50
10/03/2007 17.3 4:59:11 5:34:14 6:13:01
11/03/2006 21.0 5:13:58 5:51:38 6:30:02
12/03/2005 21.1 5:12:49 5:49:05 6:29:56
13/03/2004 17.7 5:04:28 5:39:12 6:19:17
8/03/2003         17.4 5:09:21 5:47:05 6:23:55
2/03/2002         17.1 5:08:18 5:45:14 6:20:52
3/03/2001         16.2 5:09:35 5:45:01 6:20:00
4/03/2000         22.5 5:19:22 5:54:20 6:33:11
6/03/1999         22.9 5:24:01 6:03:16 6:42:17
7/03/1998         13.7 4:58:54 5:35:20 6:05:48
1/03/1997         ??? 5:20:04 5:54:46 6:33:11
2/03/1996         17.9 5:13:36 5:47:59 6:27:20
4/03/1995         11.1 4:57:25 5:33:44 6:16:20
5/03/1994         18.8 5:09:22 5:43:47 6:29:59
6/03/1993         17.0 5:05:31 5:46:36 6:19:11
21/03/1992 17.2 5:02:18 5:37:33 6:22:22
16/03/1991 ??? 4:56:51 5:32:35 6:12:12
24/03/1990 12.7 4:55:35 5:36:23 6:16:39

h.

[The Sloth]

You wouldn't happen to have a PhD in Maths by any chance?


What you are saying is that I wouldn't have gone under 4:30 last year if the weather hadn't been so kind.
That makes 6ft Sicko status a lot further away (not that it was attainable...)

[Colin]

....and the last runner's time hasn't changed     

[hatless]

Colin, on Feb 11 2009, 11:56 PM, said:

....and the last runner's time hasn't changed     

Not quite. The analysis is done on the basis of the median of the number of starters, not finishers. The temperature does play a significant role in determining the number of non-finishers. See the graph below. This is subject to the usual cavet that cut-offs have changed over time.

Cheers,

h.

Attached Files

•  Non_finishers.jpg   44.57K   49 downloads

[FlyerUltra]

Hatless

That is a really interesting and intriguing analysis. Pisses me off that I did 4:03 in two of the hotter (21 degrees avg) years and missed out on last years run in cool conditions (15 deg avg). I hope its cool again this year

Phibes

[Colin]

What about the demographic change of the marathon runner over the last 20 yrs. I would think that despite entry qualifications (although some are not marathon times), the amount of growth (newbies) would have been over represented in the 50 percentile and above (or perhaps runners who were 'slower tha ' 4hrs over marathon).

Have you normalised for that?

[SoLucky]

Thanks hatless, well done.

[hatless]

Colin, on Feb 12 2009, 11:46 AM, said:

What about the demographic change of the marathon runner over the last 20 yrs. I would think that despite entry qualifications (although some are not marathon times), the amount of growth (newbies) would have been over represented in the 50 percentile and above (or perhaps runners who were 'slower tha ' 4hrs over marathon).

Have you normalised for that?

If you look in the original post I wrote:

"None of this analysis takes into account other factors such as ... changes in the distribution in people running the event etc."

I cannot think of a practical way of normalising for this with the data which is available. If you can, let me know, but I highly doubt you could objectively look at the demographic change in the median runner. (What other data do we have? Very noisy data from other races which themselves need to be controlled for weather, difficulty etc.)

Splitting up the data into 90s vs 00s you see that there doesn't seem to be a big change in the demographic over time. Although it is a little tough to tell because there were a few more cooler days in the 90s than the 00s.

Cheers,

h.

Attached Files

•  90s_vs_00s.jpg   41.21K   19 downloads

[Tom31]

Colin, on Feb 12 2009, 11:46 AM, said:

Have you normalised for that?

Probably the most detailed piece of analysis I've ever seen on Coolrunning and you're still not satisfied!
Thanks hatless.  Great work.  Let's hope for a cool one on the 14th.

[Colin]

Tom31, on Feb 12 2009, 04:08 PM, said:

Probably the most detailed piece of analysis I've ever seen on Coolrunning and you're still not satisfied!

Why the agro?  
I just asked a very relevant question which has applied to the marathon running demographic over the last 30 yrs.
If the marathon median times have blown out over this period, and we anecdotally accept that most of the growth came from the slower entrants (I said anecdotally..it may be incorrect) then surely 6ft times would have slowed down to...and then perhaps the temp effect may be different. (didn't say wrong)

Whether it satisfies me or not depends on whether I feel the temperature would affect me personally or not, and given my history I believe not...but my satisfaction or not has nothing to do with you...different people will get different benefits out of such an analysis...

...others may just say, "so what" ...not that I said so, btw.

[Jogger]

Quote

If the marathon median times have blown out over this period, and we anecdotally accept that most of the growth came from the slower entrants (I said anecdotally..it may be incorrect) then surely 6ft times would have slowed down too

Colin - I agree - it would be veeeerrry interesting if we had a phd maths student with a couple of spare hours to look into this.

I suspect %sub4hr %sub5hr  %sub6hr  %sub7hr  for each year should do it.

[Barefoot]

Thanks for the info hatless.

My times correspond with your data, I ran faster in the cooler years.

Any data on what the temperature may be for this years' Six-Foot?

[yeti]

hatless, on Feb 11 2009, 02:39 PM, said:

I cannot think of a practical way of normalising for this with the data which is available.

How about normalizing annual data w.r.t. to the temperature (i.e. calculate the difference between the percentile times predicted by the linear fits you've shown and the raw times) and then have a look if there is a statistically significant correlation with the year (t-test maybe) ?  Normally the proper way to look at it would be to do a multivariate regression, i.e. consider the percentile times a function of both year and temperature and then look at the relative significances of the two dependencies. However, having only 17 data points this is not going to work on this data set.  

-- EDIT --
Just went over the data myself and was rather surprised by the outcome. Lets assume a linear correlation between times and temperature, which from the data seems to look reasonable although one could argue that there is a break in slope at ~16°C. Then the temperature normalized times  appear to actually get _better_ over the years. What is even more surpising is the improvement seems to be bigger and more statistically robust for the slower end of the field than for the pointy end. The data:
75th percentile: 42sec per year  improvement, correlation coefficient (r^2) : 0.40
50th percentile: 25sec per year  improvement, correlation coefficient (r^2) : 0.21
25th percentile: 12sec per year  improvement, correlation coefficient (r^2) : 0.09

Haven't done any proper significance testing but visually and from the r^2 values it looks like the improvement in the 75th percentile times is really there whereas for the median (50th percentile) it looks questionable and 25th percentile it could well be spurious.
Btw, 42sec/year may look small, but over the 18 years from '90 to 2008 that adds up to nearly 13 minutes.

Yeti

Edited by yeti, 12 February 2009 - 09:02 PM.

[Steve M]

I know that in the 70's and 80's even overweight pensioners ran marathons in under 3 hours, but I wonder with six foot given how strategic the race is the increased information available is enabling better times. Especially at the back end of the field. If so you would expect the improvements would be flatening out.

I do love a good battle of the nerds though, I say Yeti and Hatless draw algorithms at twenty paces. Who can come up with the most mind boggling relationship to tie all these factors together.

[Colin]

I blame the sweepers and the grim reaper   

...seriously , other marathons allow infinite times whereas if you want that medal you have to get a wriggle on at the tail end...

[run2work]

Barefoot, on Feb 12 2009, 06:34 PM, said:

Thanks for the info hatless.

My times correspond with your data, I ran faster in the cooler years.

Any data on what the temperature may be for this years' Six-Foot?

Forecast for the day is light cooling drizzle, breaking into sunny skies as I cross the finish line!!

Oh sorry, I have just woken up from a lovely dream.

rtw

[CulturalStereotype]

pinch yourself R2W, you're not dreaming

http://www.bom.gov.a...N10061.shtml#CT

Forecast for Saturday
Isolated showers and thunderstorms developing. North to northwest winds.

Katoomba : A few showers.                Min:   13  Max:   22

[run2work]

Rain cooling one as one gets to top of Plu, then lightning and hail hitting ones frame, causing ...

Oh sorry, I have just woken from a night terror.

rtw

[Mister G]

... Just remember, it's a lot hotter climbing out of the Coxs up Mini Mini than it is at Mt Boyce...!

[run2work]

Mister G, on Mar 11 2009, 04:04 PM, said:

... Just remember, it's a lot hotter climbing out of the Coxs up Mini Mini than it is at Mt Boyce...!

yes, but remember the refreshing dip in the river, and the wonderful drink station positioned there.

rtw

[nando]

So, if I've got it right, this years ave temp was 17.6, which makes it a fairly normal year.

I thought the conditions on Saturday were excellent, but it looks like last year was even kinder, with the average 2 degrees cooler.

Edited by nando, 16 March 2009 - 01:32 PM.