Training based on TSS instead of HRS



acoggan said:
rchung said:
I'm not sure how many modern statisticians think that this is a good way to account for interactions.
Trust me, I've dealt with plenty of statisticians, and every one that I have ever met has modeled interactions that way. It's also how the commonly accepted software packages - SAS, SPSS, etc. - routinely deal with interaction effects.
Oh dear. That's exactly why I put "modern" in "modern statisticians".
 
acoggan said:
Well sure...and given your penchant for math, I'm a bit surprised that you've only recently jumped on the bandwagon with the rest of us. ;)
I have to admit I understimated the power data analysis area (okay, I understood kJ, and power-based zones, and pacing for steady power instead of HR. Not very exciting). There's obviously been substantial progress in the past 5 years or so (at least in cycling), a lot of it you've obviously been responsible for.
 
djconnel said:
I have to admit I understimated the power data analysis area (okay, I understood kJ, and power-based zones, and pacing for steady power instead of HR. Not very exciting). There's obviously been substantial progress in the past 5 years or so (at least in cycling), a lot of it you've obviously been responsible for.
Andy is the leader for sure. Re diff. load measures: I have about three years worth of data split into indoor and outdoor rides and using load measures of: plain kJ, std. TSS(^2), linear.TSS (^2) and linear.TSS(^4).

I liked the linear.TSS(^2) porridge best of all. But perhaps it is just down to taste. After doing 5-6 months of indoor training each year, those easy pedalling/coasting ride sections just beg to be cleaned up. Difference is typically 10-15% IIRC over the week.
 
djconnel said:
I have to admit I understimated the power data analysis area (okay, I understood kJ, and power-based zones, and pacing for steady power instead of HR. Not very exciting). There's obviously been substantial progress in the past 5 years or so (at least in cycling), a lot of it you've obviously been responsible for.
here's another "leading indicator" :) Congrats Andrew.

"Andy Coggan, PhD, has been named a finalist for the USOC "Doc" Counsilman
Award. The award is given to coaches who have developed methods for
integrating sport science into coaching.

Dr. Coggan's work in developing mathematical modeling for using power as a
basis for training.

Dr. Coggan has been a co-author of a book on power based training and
lectured both for USA Cycling and at international conferences.

USA Cycling congratulates both men [Andy Sparks was nominated for a USOC
Coaching award for his work with Sarah Hammer] for their respective honors.

The awards will be presented May 3 in Washington, D.C. at the President's
Council on Physical Fitness and Sports Award luncheon."
 
acoggan said:
Thanks Rick - where did you see that, BTW?
Over on ST a while ago and I forgot to comment on there. No I'm not a coach nor even a reasonable fascimile of one ;0)
 
rmur17 said:
Over on ST a while ago and I forgot to comment on there.

Ah - I was curious because that sounds like a USAC announcement, but I've yet to see one from them.

No I'm not a coach nor even a reasonable fascimile of one ;0)

:D
 
acoggan said:
Ah - I was curious because that sounds like a USAC announcement, but I've yet to see one from them.



:D
uhm ... a little bird just told me it was indeed from a USAC newsletter .. different version perhaps?

I foresee your acceptance speech: "Beyond kJ" :D
 
rmur17 said:
uhm ... a little bird just told me it was indeed from a USAC newsletter .. different version perhaps?

That explains it: not being a coach, I don't get the newsletter. ;)

(Actually, I read the USACCA list online, so don't get the newsletter.)

rmur17 said:
I foresee your acceptance speech: "Beyond kJ" :D

:)
 
rmur17 said:
Andy is the leader for sure. Re diff. load measures: I have about three years worth of data split into indoor and outdoor rides and using load measures of: plain kJ, std. TSS(^2), linear.TSS (^2) and linear.TSS(^4).

I liked the linear.TSS(^2) porridge best of all. But perhaps it is just down to taste. After doing 5-6 months of indoor training each year, those easy pedalling/coasting ride sections just beg to be cleaned up. Difference is typically 10-15% IIRC over the week.
It seems the conflict is to between getting proper credit for big efforts over threshold, versus losing adequate credit for long efforts below threshold.

One approach to this would be hyperbolic sine ( sinh(x) = (exp(x) - exp(-x) / 2), where the "power" in effect increases with increasing power above threshold. For example, you could say:
TSS = (100/hr) integral { [ K sinh(P/FTP) ]^2 dt }

Attractive, but unfortunately I looked at Andrew's lactate level data from:
http://www.midweekclub.ca/articles/coggan.pdf
and those data ARE really well fit by P^4, not sinh^2.

So it seems to me what one wants is to consider multiple sources of stress. There's one source of stress from blood lactate, another from doing work. At some point, it stops being easier to climb a hill slower. If I do a 1000 meter climb in 8 hours, it's no easier than doing it in 4 hours. Work is work. Of course, if I do it in 1 hour, that gets harder. Even w/o elevated lactate, I get muscular fatigue, more than if I were laying in bed reading.

So perhaps the best approach is neither ^1, ^2, ^3, or ^4 but a combination of P^4 proportional fatigue and good old fashioned kJ:

TSS(t + dt) = TSS(t) + (100/hr) [ a (P/FTP) + (1 - a) (P/FTP)^4 ]

for some a. For example, if a = 0.5:

TSS(t + dt) = TSS(t) + (50/hr) [ (P/FTP) + (P/FTP)^4 ]

Of course, you can still add in a term like I suggested before, for the effect of accumulated ride fatigue.

Dan
 
frenchyge said:
Not really, that's why I snip out any significant periods of noodling within an otherwise productive workout. I don't have any 20min descents to worry about here, but we do occassionally have to stop for a coal train. ;)
Good approach to snip out BS sections- but how many people do you think don't do that.
 
acoggan said:
It's called "training stress score" and not "training performance score" for a reason.
Exactly- Descending for 20 min has nothing to do with either stress or performance.
 
gvanwagner said:
Good approach to snip out BS sections- but how many people do you think don't do that.

Not many, because it usually has minimal impact on the TSS. For example, after snipping out the ~15 min of coasting that my wife did during a recent ~2.5 h ride, here TSS only changed by ~1%.
 
acoggan said:
Not many, because it usually has minimal impact on the TSS. For example, after snipping out the ~15 min of coasting that my wife did during a recent ~2.5 h ride, here TSS only changed by ~1%.
she wasn't riding hard enough the other 2.25 hrs :D :D :D
 
rmur17 said:
she wasn't riding hard enough the other 2.25 hrs :D :D :D
That shouldn't matter: the effect should be fractionally the same no matter what the rest of her ride. 1/4 hour of coasting added to 9/4 hours of riding reduces NP by a factor (10/9)**(1/4), and therefore reduces (NP/FTP)^2 by (10/9)^(1/2), and therefore increases TSS by (10/9)^(1/2) (because duration is increased by a factor 10/9). This is a 5.4% increase in TSS. I'm not sure how Andrew got only a 1% difference: the math is straightforward, unless the coasting intervals were so short the 30-second rounding was affected.

Dan
 
acoggan said:
Not many, because it usually has minimal impact on the TSS. For example, after snipping out the ~15 min of coasting that my wife did during a recent ~2.5 h ride, here TSS only changed by ~1%.
I don't agree with that at all, the difference can range anywhere from minimal (USUALLY my rides are changed by less then 1%) to 25% depending on the workout. Im not advocating that we change TSS but like any tool, people using it to such a large degree need to be aware of its tendencies. Especially those files most likely to trigger RATSS. Besides the workout files that show it I see it in races with long descents and ones where the cool down is included. When you have rides that are 20%+ coasting or super low power in chunks of more then 5 min then then a person should look at it critically.

Greg
 
djconnel said:
This is a 5.4% increase in TSS. I'm not sure how Andrew got only a 1% difference: the math is straightforward, unless the coasting intervals were so short the 30-second rounding was affected.
I suspect that's it: I believe Andy lives in a pretty flat place so he and his wife probably rarely see rides with longish coasting intervals.

A deeper issue (that I usually try to ignore) is that the entire implementation of TSB depends on the additivity of TSS.
 
djconnel said:
That shouldn't matter: the effect should be fractionally the same no matter what the rest of her ride. 1/4 hour of coasting added to 9/4 hours of riding reduces NP by a factor (10/9)**(1/4), and therefore reduces (NP/FTP)^2 by (10/9)^(1/2), and therefore increases TSS by (10/9)^(1/2) (because duration is increased by a factor 10/9). This is a 5.4% increase in TSS. I'm not sure how Andrew got only a 1% difference: the math is straightforward, unless the coasting intervals were so short the 30-second rounding was affected.

Dan
okay 1% just sounded small to me. I'm not much on working out the gory details.