Kenenisa in 26:17 Form Could Run 2:00:36? So Say MIT Guys

said88 wrote:

OK, thanks for this observation!

But then really - in my opinion - they shouldn't include the Mile in their calculations at all.

But I'm also not convinced with other results.

3:26.00 (1500m) + 26:20 (10000m) gives 12:40.23 for 5000m.

Shouldn't this wonder-runner easily be capable breaking the WR?

I also checked some other combinations (distances and/or times) and I was - as said - not very impressed.

But for sure this isn't something like an "scientific analysis" of this calculator and - again - I have not read their paper.

I agree, and I raised this with the creators in another thread which I believe was deleted.

Again, of course you are completely correct that runner should be able to smash the 5k WR. Unfortunately, I think the 10k suffers from the same issues as the mile. It is sparingly run and so the predictions using it are quite innaccurate.

On a related note, a small rule of thumb I have come up with considering your point is this:

1.35 x 1500m + 0.3 x 10,000m = 5000m

So with your examples it predicts 12:32 for the 5000m.

Also it suggests Bekele could run 3:30 around his WR shape.

Precious Roy wrote:

They are two completely different events. It is almost like saying Novak Djokovic beat Federer in straight sets, so he could shoot a 74 at the Masters. The marathon is all about fueling. Your flat out track speed does not mean anything in the marathon. Mo Farah crashed and burned in his Marathon debut. Tadese can't break 2:10. Kimetto and Kipsang are purely road racers with almost no track experience. But then you have Kipchoge who was a beast at 5k and never did many 10ks before moving up to the marathon. Thus, any attempt at finding an objective formula to predict a marathon performance is just silly. Each athlete is different. You just cannot tell who will excel at the marathon until they run one. It is a totally different event.

Why do you need to assert that the marathon and 10k are "totally" different events?!

They are not totally different events. The marathon and shot put are totally different. The 100m and marathon are very different events. Running is key in both.

I've raced in marathons and 10ks. They are a lot alike. You've got to burn fuel efficiently while running in both. More so in the marathon because it's over 4 times as long.

If you can run a 10 k in 27 min then you can probably run a marathon in 2:11 + or - 4 %.

But for a 10 sec 100m guy we can't predict a marathon performance within + or - 4 % with much confidence at all. They are very different events... not totally different because they are both running events.

But then - sorry - I think the calculator is not very helpful at all. When everytime I want to use it I have to think why for this situation it doesn't work, well...

When a simple equation as the one you have listed is giving better results as the calculator these guys have created, then I think there attempt maybe was not very well thought-out?

I agree that Bekele in 2004 probably could have come very close to 3:30.

I think we all agree that 10000 and Half-Marathon are quite well connected events. However, if I enter 00:26:30 for 10000m as single past performance, I get 00:57:02 for HM that is totally irealistic...

I also tried with two performances, and it does not match with Peronnet and Thibaut formula.

You probably know that the Canadian physiologists Peronnet and Thibault have found (1,2,3) that for each runner a linear relation exists between the %vVO2Max and ln(t_lim/t_vVO2), with :

- t_lim the time that could be sustained by the runner at the velocity corresponding to %vVO2Max considered ;

- t_vVO2 : the time that could be sustained by the runner at vVO2Max

The slope of this linear relashionship is the "Aerobic Edurance Index" (AEI) of the runner, and indicates the decrease of speed as the duration of the event increases (from 5k, to 10k, to HM for instance). So, the less negative the slope is, the more the runner is able to maintain speed as the duration of the event increases, the best his aerobic endurance is.

Typically, a very good AEI is about ~ –4.5 to –5 (Paula RADCLIFFE is –4.7 for instance). (4) The level of the AEI has been found to be largely connected with the proportion of Slow-Twitch vs Fast-Twitch fibers (the more ST, the best the AEI is). Of course, provided the performances you use to compute the AEI of a specific runner have been optimized (same absolute level, and in both cases similar conditions, and optimal training for the distance).

Now, when you consider a runner with PBs of 13:11 at 5k and 27:23 at 10k, so with a very good AEI of –4.9, based on Peronnet's formula a HM time of 1h00'17 could be predicted for this runner, (4) that is already more realistic than the value provided by your algorithm...

But now, if another runner has the same PB in 10k of 27:23, but is faster at the 5k with a 13:05 PB, that means he looses more rapidly speed as the duration of the even increases than the first runner, and so his AEI is poorer (–5.75 exactly). Thus, his HM performance should be WORSE than the one of the first runner, and the second runner have a less good aerobic endurance, but with your algorithm we find THE OPPOSITE...

This is another point that is unrealistic with your modeling.

(1) Peronnet, F. & Thibault, G. (1987). Analyse physiologique de la performance en course à pied, révision du modèle hyperbolique. J Physiol. 82, 52-60

(2) Peronnet, F. & Thibault, G. (1989). Mathematical analysis of running performance and world running records. J Appl Physiol. 67, 453-465

(3) you may also see (sorry, it is in French) :

http://www.fredericgrappe.com/wp-content/uploads/2011/10/Indice%20d%20endurance%20et%20profil%20de%20performance%20en%20cyclisme%20%20etude%20preliminaire.pdf

(4) The AEI could be determined provided you have at least two optimized performances from this web site (also in French...) :

https://plans-entrainement.net/estimateur-de-performances-calculateur-dallures/

BBen wrote:

I think we all agree that 10000 and Half-Marathon are quite well connected events. However, if I enter 00:26:30 for 10000m as single past performance, I get 00:57:02 for HM that is totally irealistic...

Bekele's WR times give a 1:58:39 Marathon time...

And people talk if a sub 2 is POSSIBLE!

At least I have learnd now that I really don't have to read the long paper.

Just for mentioning: using Peronnet and Thibault formula for Paula RADCLIFFE, to guess to which performance in the HM her 2:15:25 would correspond, FOR HER.

If I take two performances close in time, and both "optimized" (presumably both very well prepared), I can determine Paula's AEI:

- If you take her WR in HM dating of Autumn 2001 of 1:06:47, and you couple with her 2:18:56 in London the following spring, you find and AEI of –4.7. (1)

- If you take her PB of 30:01 in 10000m from August 2002 (EuropeanC), and you couple it with her following 2:17:12 in Chicago a few months later, you also find the SAME AEI of –4.7...

Then, you can try to determines which HM time is connected to her 2:15:25 through an AEI of –4.7, you find : 1:05:05.

Which is quite realistic... And also this is what made me wonder in another thread lately why Florence KIPLAGAT has not ran at least a 2:17/2:18 marathon yet... and conclude that her AEI should be rather poor, meaning she is probably too much "FT" for the marathon.

(1) still using :

https://plans-entrainement.net/estimateur-de-performances-calculateur-dallures/

For those you can read french, you may also like to read this article about Peronnet and Thibault formula (with an interview of Guy Thibault):

https://www.dropbox.com/s/k018a5jqs2w7eeb/Sport%20et%20Vie%20Endurance.pdf?dl=0

BBen wrote:

Now, when you consider a runner with PBs of 13:11 at 5k and 27:23 at 10k, so with a very good AEI of –4.9, based on Peronnet's formula a HM time of 1h00'17 could be predicted for this runner, (4) that is already more realistic than the value provided by your algorithm...

But now, if another runner has the same PB in 10k of 27:23, but is faster at the 5k with a 13:05 PB, that means he looses more rapidly speed as the duration of the even increases than the first runner, and so his AEI is poorer (–5.75 exactly). Thus, his HM performance should be WORSE than the one of the first runner, and the second runner have a less good aerobic endurance, but with your algorithm we find THE OPPOSITE...

This is another point that is unrealistic with your modeling.

Absolutely correct! For sure a faster 5000m time should result in a slower HM time.

Well the predictions are based on the results of thousands of runners. It is a stastically sound way of doing things, and they have taken a scientific approach to the matter.

The problem is the data screening and the quality of the data itself. Their method seems OK.

My formula is simply a rule of thumb based on my observations.

I'd say consider their work a beta. That said, their predictions using the 400/800/1500/5k are OK.

Significantly better than Ventolin's.

Metric Miler wrote:

Well the predictions are based on the results of thousands of runners. It is a stastically sound way of doing things, and they have taken a scientific approach to the matter.

But for sure you have to consider such things that specific events are run (significantly) less often then others.

Sorry, but when the outcome is like it is in this case, then all the work was not very well thought-out - at least I have this impression.

I have read "scientific" articles in very well known german newspapers which - for example - said that with 95% (I better don't ask them how they got this number) security the absolute limit for humans in the 10000m is 26:38 - well, the article was published before (but not much) summer of 1996.

Or that women will reach the men's level in some events in something like 50 years. Based of some analysis of the record progressions of the last decades. Well, in such results definitely also is some sort of feminism involved, but especially a very big amount of not beeing very well informed about the history of athletics. But this is needed deeply to come to good results in such questions.

I can see that it is reasonable to get the impression that if part of the research is inaccurate, then it is all inaccurate. However this is not the case and not always true. The method can be sound and the algorithm developed could yield accurate results when provided with high quality data.

The issue with this kind of work, predicting athletic performance, is that the data is in itself unreliable. Using large amounts of that data does not do enough to negate effects that are difficult to account for, such as weather conditions, the tactical nature of a race, injuries and illnesses affecting a performance etc.

If we had a list of 'all out', perfect condition races for individuals across a range of events, then producing a formula or algorithm would be trivial.

I personally take the approach of quality over quantity with my little 'rule of thumbs' I have developed (and are a work in progress) to use as training aids.

I used data collected about individuals where performances are close together and maximal efforts in similar conditions (where possible).

I found the results to be quite accurate, especially for the 400/800/1500 one I have developed.

These are not extremely accurate; they are to give ball park figures.

As some others have shown, making 5+ significant figure predictions is silly, unjustifiable and generally poor practice. The number of variables is too large.

bump

bump