Doping -- Le Monde editorial

"benjo maso" wrote in message
...

"Daniel Connelly" wrote

Ed-D wrote:

the article notes that the average speed of stage winners this
year is the fastest on record, even faster than in 1997, the year
preceding the Festina affair, when EPO use was presumably widespread.

2003 : 3427 km
1997 : 3942 km

Average speed 1997: 39,273
2003: 40,956

A difference of almost 1,7 km/h. Only a little bit more than between 1997
and 1962 (4272 km, average speed 37,304).

http://mywebpage.netscape.com/rechung/temp/tdf.png

the article notes that the average speed of stage winners this
year is the fastest on record, even faster than in 1997, the year
preceding the Festina affair, when EPO use was presumably widespread.

2003 : 3427 km
1997 : 3942 km

Average speed 1997: 39,273
2003: 40,956

A difference of almost 1,7 km/h. Only a little bit more than between
1997
and 1962 (4272 km, average speed 37,304).

http://mywebpage.netscape.com/rechung/temp/tdf.png


Very informative plot, A couple of points:

1. I think Le Monde (the newspaper, not the rider) was discussing the
average speed of all the stage winners, rather than simply the single GC
winner. This would spread the measurement over many riders, rather than just
focusing on a single one. Unfortunately, the article didn't mention whether
this average was distance-weighted, but this might not make a great deal of
difference.

2a. It would be nice to see two plots: the stage-winner average speed vs.
Tour length and (2) stage-winner average speed vs. year, which might address
the allegations more directly.

2b. In fact (and this would be the best solution) you could "correct" the
average speed for tour length, and then plot this corrected speed vs. year.
Then you would be looking at how the average speed of stage winners has
changed over the years, but taking into account the varying Tour lengths. If
the observed increase in average speed (which, as you point out, is pretty
small to begin with) were entirely due to the decrease in Tour length, this
plot would be flat, i.e., without any trend, and would effectively refute
the Le Monde allegation.

It might be tricky if you found, say, a rather small trend towards
increasing speeds with time. You would have to decide (a) if this trend was
statistically significant/or (b) if it might be due to yet other factors,
such as reduction in bike weight over the years.

Good material for a statistics term paper (and no, I'm not a student!)

-------------------

"AMG" wrote in message
ink.net...

http://mywebpage.netscape.com/rechung/temp/tdf.png


Very informative plot, A couple of points:

1. I think Le Monde (the newspaper, not the rider) was discussing the
average speed of all the stage winners, rather than simply the single GC
winner.

They were, but I don't have that info for all of the Tours. For the Tours
for which I *do* have that info, the "moyenne vitesse de la course" is
consistently about 0.2 kph higher than the "moyenne vitesse du vainqeur."
That's small enough not to make much difference, at least with respect to
the Le Monde editorial.

2b. In fact (and this would be the best solution) you could "correct" the
average speed for tour length, and then plot this corrected speed vs.
year.
Then you would be looking at how the average speed of stage winners has
changed over the years, but taking into account the varying Tour lengths.
If
the observed increase in average speed (which, as you point out, is pretty
small to begin with) were entirely due to the decrease in Tour length,
this
plot would be flat, i.e., without any trend, and would effectively refute
the Le Monde allegation.

Well, what you're talking about is something related to a residual plot. I
was trying to stay away from that stuff because most people wouldn't
understand what that is, but I've put one up at
http://mywebpage.netscape.com/rechung/temp/tdf2.png

This is based on a linear model of speed based only on a single predictor,
i.e., distance. I'm not really keen on linear models but they're often handy
for doing cursory stuff like this. BTW, I've added a little line to the
original plot that shows the linear regression line.

I posted the plot without any interpretation because I'm sort of interested
in what people see in it. Think of it as sort of a Rorschach inkblot test.
However, since you've been kind enough to look at it, I'll tell you what I
see: one single variable (distance) seems to be a moderately good predictor
of average speed. Without knowing anything else about the race (the weather,
the amount of climbing, the tactical or strategic situation) except total
distance, the fit is moderately good. The 2003 race was the fastest on
record, but it was also one of the shortest. The 2003 race has a positive
residual (i.e., it was faster than would be predicted by distance alone) but
not remakably so. The most aberrant races were the 1947 and 1989 races,
which were much slower than one would have expected from a model of speed
based on distance alone. In addition, the recent races are almost a third
shorter than the races of 50 years ago which, in this simplified view of the
world, happens to explain a fair amount about the increasing speeds.

In case you think I've done something wrong, I've posted the data underlying
these graphs at
http://mywebpage.netscape.com/rechung/temp/tdf.csv
in comma-separated-value text format.

"Robert Chung" wrote in message
...
The most aberrant races were the 1947 and 1989 races,

Sorry. Of the 57 races since WWII, the most aberrant races (in terms of the
size of their residuals) were the 1947 and 1973 races.

Yeah, Rush Limbaugh...he's a reliable source. I love republicans;
always crying liberalism in the media. Too bad they control most
major media sources, which is why you hear so little about anything
illegal regarding republicans, and everything sexual about Clinton.

Now let's take this discussion off RBR.
Jay Hill wrote in message ...
Ed-D wrote:
Not sure what that is supposed to mean, but it's common knowledge
that the
Times is left-of-center.

Common knowledge among whom? Dittoheads? So you think every one of
the thousands of employees at the Times takes a little liberal truth
test and swears fealty to Jesse Jackson & Hillary Clinton?

Would not the total elevation gain over the entire Tour, or at least
over the major mountain stages, play a big role in average speed? Or
is it the case that that has remained fairly constant from year-to-year,
or at least its ratio to total tour distance?

Best, Peter

This is based on a linear model of speed based only on a single
predictor,
i.e., distance. I'm not really keen on linear models but they're often
handy
for doing cursory stuff like this. BTW, I've added a little line to the
original plot that shows the linear regression line.


Would not the total elevation gain over the entire Tour, or at least
over the major mountain stages, play a big role in average speed? Or
is it the case that that has remained fairly constant from year-to-year,
or at least its ratio to total tour distance?

I agree completely. The Le Monde editorial implicitly assumes that all Tours
are equal, in the sense that if a particular year's Tour is 20% longer than
the average, it will have 20% more mountain than the average, "mountain"
including all the elements that cause riders to slow down. So a detailed
knowledge of every year's course -- as well as a lot of time -- would be
necessary for a thorough analysis.

Re the actual data (average speed of the GC winners, 1947-2003), a little
calculation indicates that for each additional 100 kilometers of Tour
length, the speed decreases by 0.41 km/h. This is an average figure over the
above period. There are plenty of deviations from the average, and
variations in the mountain fraction could be one reason for these
deviations.

Overall, average speed was in the low 30s from 1947 until about 1955, when
it increased to the mid 30s. It started climbing unevenly around 1975 and
then more steadily after about 1985. Nearly constant from 1995-1997 at
around 39 km/h, it climbed, with a couple of small wiggles, to its all-time
high in 2003, BUT:

If the speed is corrected to take account of the effect of Tour length, a
different picture emerges. The period 1955-1965 now appears as a broad
plateau, after which the speed declines during 1965-1975 and increases, with
some large fluctuations, from 1980 to about 1991. After this it is more or
less stable all the way through 2003, which is no longer the fastest Tour on
record. The two fastest tours would now be 1992 and 1998, and the LA years,
1999-2003, show a steady decline except for 2003, which is now the second
fastest LA tour, after 1999.

A real Tour historian (or a pharmacist) might be able to explain some of the
wiggles on the plots, and this would be useful. But it is interesting that
the rocket-like trend in GC winner's speed shown by the "raw" data is not
really present when the speed is adjusted to take account of tour length.
And Robert's plot makes the length-speed relationship pretty clear.

I would be happy to post the plots of speed and length-adjusted speed but I
don't have a web page to do it to at the moment, so if someone else wants to
do it I'll be pleased to send a file (it was done in Excel). Food for
thought...

AMG
---------