SRM vs CPL and Ferrari estimates

frenchfighter

mulletmaster

Given the SRM has 1% tolerances that is good going, not far out at all to be honest.

davidof

Were the estimates made prior to the SRM data being available or after? That is, it always seems easier to backtest than to predict.

morstar

Lies, damn lies and statistics.
What is sample set, criteria, etc. etc.
Random selections of data can be used to prove 101% of anything and everything. A stopped clock is right twice a day.

Taking a leap of faith and accepting the data at face value...
What does it then prove? Horner and Froome are cheats, Contactor isn't? We should buy euros or Syria should be invaded.

gpreeves

morstar wrote:

Lies, damn lies and statistics.
What is sample set, criteria, etc. etc.
Random selections of data can be used to prove 101% of anything and everything. A stopped clock is right twice a day.

Exactly, 5 observations do nothing to prove the accuracy or inaccuracy of these models - it is too small a sample. The accuracy does appear to be impressive, but we have no idea if these were chosen because they fit the observed outputs so neatly.

RichN95.

This is a link Blazing posted on a another thread showing a comparison of various methods with the SRM values for seven climbs by Horner in the 2010 Tour:

http://www.fillarifoorumi.fi/forum/show ... ost2017096

Some of the estimates are quite accurate, but equally a lot are 4-10% off. (And bear in mind these are all for the same rider so the number of variables is reduced). And on 6 of the 7 climbs the estimates are always too high.

Now, that's not an issue if you're just doing this for a bit fun, but a sizable problem if you want to use it as the basis of an accusation. In Vayer's (completely unsupported) criteria 10% is the difference between being above suspicious and 'miraculous'.

No_Ta_Doctor

RichN95 wrote:

This is a link Blazing posted on a another thread showing a comparison of various methods with the SRM values for seven climbs by Horner in the 2010 Tour:

http://www.fillarifoorumi.fi/forum/show ... ost2017096

Some of the estimates are quite accurate, but equally a lot are 4-10% off. (And bear in mind these are all for the same rider so the number of variables is reduced). And on 6 of the 7 climbs the estimates are always too high.

Now, that's not an issue if you're just doing this for a bit fun, but a sizable problem if you want to use it as the basis of an accusation. In Vayer's (completely unsupported) criteria 10% is the difference between being above suspicious and 'miraculous'.

To be fair to Vayer, I think most errors in the figures he reports come from the fact that it's actually really hard to write in crayon on the back of a fag packet and it's not always easy to read afterwards.