What if I want to know what power I can expect for 8 min in the middle of a race, i.e. am I going to make the selection on a particular hill? I could give a 'maximal' effort but it wouldn't be the same maximal effort I might give if I was fully rested.

Not sure I see the value in arguing about the precision of a fuzzy curve. Just because it's drawn with a sharp thin line doesn't mean it's not fuzzy.

MAP contribution in Peronnet matches FTP contribution in WKO4
AWC contribution in Peronnet matches FRC contribution in WKO4

Its been discussed before on wattage* and here. The WKO4 model components were digitised and compared with Peronnet. The spreadsheet is here: https://docs.google.com/...pi-tkFFu4ewDj2VyfZQ/

The only way out here for Coggan is if the data isn't from the WKO4 model (it was digitized from a WKO4 screenshot **).

Mark

* https://groups.google.com/...feZQt-o/ja_D7_urAQAJ
** https://drive.google.com/...8TK/view?usp=sharing

Hi Greg,

That's an interesting problem that is being addressed by the W'bal (and Xert MPA) models.

You still need to know what your CP and W' are (or FTP/FRC if you prefer that terminology). And they need to be pretty accurate, so the problems we have been discussing here don't go away.

But then once they are known, you can use the W'bal model to see how W' goes up and down through a ride. At any given point you will have used or replenished W' to some degree -- and that is what the W'bal model does.

Then to know what your power is at 8mins into your race you can calculate it using the W' remaining in your tank at that point in time.If you have a garmin watch there are some apps you can download to see this in real time. Its pretty neat.

One word of caution, the W'bal model is quite a hot topic for research and the recovery kinetics are especially interesting. A recent paper showed that method for calculating W' recovery time for elite cyclists is quite different to the typical competitive cyclist. So at present its a work in progress -- for some it works well, for others less so.

Hope that's useful.

Mark

Disclaimer: I don't actually know anything behind the WKO4 model, and everything is based on intuition and what has been said here.

The basic answer to why the model might be underestimating to such a degree is that even though you have data that shows your mFTP is higher than what is predicted, these additional data points don't contribute enough to the error function that the model is attempting to minimize for it to move the curve up for these time periods.

A longer answer is that since OLS is used and the PD curve doesn't seem linear (even with the log scale), each data point that you have (t, P) is mapped in some nonlinear way to a different space, and a line is fit to the data in this new space, with the slope and intercept chosen such that the sum of squares error is minimized. Predictions can then be made by reversing the nonlinear map to the original space (time, power). The key is point is that each data point is given the same weight in the error function, so if you have N data points and N is large, an additional data point likely won't move the curve very much (if the curve moves toward this new point, the cumulative error on the previous points due to the move will outweigh the benefit moving to the new point). If N is small, an additional point may have significant influence on the curve (but again, if moving the curve towards the new point increases error a lot for the old points, the model won't do it).

Additionally, it is possible that due to the mapping, the data points at long time periods are "squished" so that differences with the PD curve mean less than data points at short time periods (to see this, consider a log transformation, log(500) - log(400) < log(200) - log(100)).

---

I like analyzing things - http://engineeringfitness.org

If someone had had the presence of mind to actually patent the false claim, they would be enormously rich by now.

I think a Damped Least Squares algorithm is actually being used in WKO4. Coggan has previously stated they use a custom build Levenberg-Marquardt solver. His earlier post is likely in error, or the code was changed recently.

Not that this in any way addresses your main point about weighting.

stumpyx13, good explanation!

What actually this thread is all about, can be summarized in the quote by Box: "All models are wrong, but some are useful".

In this case, as a scientist, Dr. Coggan religiously defends the accuracy of his model (and I totally understand why he's doing it), and from statistical perspective, his model likely is the scientifically best there is. It would be really cool if WKO4 printed out also some overall accuracy estimates of the model for each athlete - then you maybe could see that you thought last ride suggested your FTP being 20W higher, but model hardly moved while accuracy of the curve fit actually slightly decreased ;)

However, what many other posters have raised here, the other important thing is the usefulness of the model. The model can be scientifically accurate to the maximum, but its "value" for "normal people" comes out of its application.

As a personal example, to me, any model is useful only if it can predict something (future performance, areas needed for improvement, training focus necessary etc) with notably better probability and/or accuracy than my (educated) guess (educated because I've trained long enough to know myself well enough). If it can't, I see no point of paying attention to such model except for bragging how much my curve is higher or lower than anyone else's.

---

----------------------------
Need more W/CdA.

No-one is criticising the model. It is the data that is used to fit which is at issue.

If OLS is causing the model to come in 12w low after being fed 60MMP data then you might consider another approach.

Sorry if I am being obtuse but I don't entirely follow.

Are you saying that the data points are within whatever margin of error the model uses? (might be nice to know)

....and I'm curious about what N might be....ie how many times to you have been above the prediction to make it count?

Here is the update curve as promised. The actual is below the model from ~1:20 to 3:00min and from ~3:20-4:20.

.....will improving those improve the approximate 10W gap at 1:03:30?

(I am not WORRIED about that and it does not affect my training....though it may make me start out too conservatively in a race of that duration....I just want to try too understand how the model works).

As a cycling coach the purpose of a well constructed power duration curve is huge. A it helps us understand the demands of the event, we can assess where the rider is at and do some training and try and match the current curve to the one we want in preparation.

As far as prediction. That is what some people call the Art of Coaching. I call it BS!

---

Hamish Ferguson: Cycling Coach

I don't understand what you are trying to say here. The WKO-model does not appear to be very useful for measuring where the rider is at, and I think even Andy suggests just measuring performance instead of using a model to do this.

No idea why you need to bring up BS.

Having had a play around with the FTP-FRC curves in WKO4 as well as the Peronnet model data that Mark posted a link to a few pages back, I’d say performing a max effort in the 1-3 min range is unlikely to have much of an effect at all on the PD-model in the 1 hour region.

I suspect that any effect on your aerobic contribution would be minimal because that would be dominated by the numerous other MMP datapoints, and whilst it may benefit your calculated FRC, the effect of that on PD at the 1-hour timepoint, or indeed mFTP, would be absolutely minimal.

Further to your point about the underestimation at 60 mins, amongst the (extremely) small differences that I noticed between the WKO4 and Peronnet models was some extra smoothing protocol applied around TTE to ease into the drop-off in aerobic power at longer durations. You can see this part of the model begin to dominate at the inflexion point in your PD-curve just before the 1:00 hr mark.

For you, me, and I suspect a lot of other riders, the lack of hard efforts at longer durations (e.g. 2+ hours) means part of the model kicks is likely to kick in a lot earlier than is perhaps representative of your capabilities and the extra smoothing effect applied appears to exacerbate the under-representation of MMP around the longest “good” datapoint for maximal effort. As such, I would suspect performing a longer hard-effort (e.g. 2-hours) might help mitigate some of this, by pushing this part of the model away from your point of interest (i.e. 60 min) especially given how flat the PD-curve is in this region.

In layman’s terms, you need good MMP datapoints (i.e. proper maximal efforts) both before and after your point of interest (in this case MMP60) for the model to work at its best.

I’m aware that this is well into the territory of chasing data just to make the curve represent reality and the practical value of that and whether it’s worth sacrificing other training regimens is up to you.

To address my own point, I'm thinking about giving this a try to see the effect.

I had a little look on TrainerRoad and found a reasonable candidate to put in a hard two-hours: Whiteside. It's a warmup, followed by a nice-and-simple, 5x 20 mins @ 95% FTP with 5 min rests; I figure breaking it down to 20 min chunks will help and I can power through the rests as appropriate.

I've no doubt that people will think I'm wasting my time but a hard-effort is still a good workout and seeing as its the off-season I'm not too fussed about overall training trajectory at the moment. I don't think it'll have a big effect on FTP but should push TTE up a bit.

If nothing else it will be a great workout !

"Great" or "stupid"... I'm not sure! I did a 4x 20 min @90-95 FTP not so long ago and that was bad enough.

Either way, it'll give me an excuse to have double-desserts this evening under the guise of carb-loading... and that's what winter training is really all about.

Which FTP was that? Old FTP a power you could maintain for approx an hour or for a 40k TT, or new FTP which is a lower power that you can only maintain for 35 to 45 minutes?

Explaining why TTE is so imprecise and misleading it tells the OP they can only maintain their underestimated FTP for 45 minutes when they have maintained that FTP for 60 minutes doesn't make it useful.

Bold claims made about TTE here by TrainingPeaks - They should make it clear how imprecise TTE is.
https://www.trainingpeaks.com/...tion-metric-in-wko4/

Yes, I don't think you understand.

A well developed power duration curve is essential to the goal setting process of determining the demands of the key events. I did this for the recent 7 day Tour of Southland modelling the power to weight demands of each stage for those who wanted to win, those who wanted to place and those who wanted to survive.

Then compared this against where the riders were at leading in and where they were at in their preparation. Then after to see how well we did for this.

Now we shift to a focus on NZ Road Nationals where the power duration curve will look different in a one day race to a 7 day stage race.

Andy pointed out in the first post that the OP's power curve was not well developed with minimal data. Some always start a training block after a rest period with a block of testing to see where riders are at.

The other benefit of the power duration curve is iLevels giving you intervals at the right intensity and right duration for your power duration curve. So better have a well developed power duration curve.

Yes, I call BS. Lot of people happy to sell you a number, that has no physiological basis. Lot of maths wizz's out there, very few decent coaches and sport scientists who will challenge those numbers. Thanks to guys like Alex and Andy who do!

---

Hamish Ferguson: Cycling Coach

No offence, but your posts come across as somewhat arrogant.

Firstly, I acknowledged the limitations of this particular 90-day dataset in my own OP. I could have easily posted one from data encompassing the whole season that is more comprehensive and more accurate, but it wouldn't have helped to illustrate the topic at hand.


Secondly, you seem to be missing the main point of this thread, which is to discuss the statistical methods underpinning these models and the potential inaccuracies they can induce in the derived metrics. If you're not willing or able to acknowledge these as a coach, then you could easily lead your athletes down an inappropriate path.

And Andy has given you a good explanation why your chart looks like it does.

Perhaps you should post the chart of the whole season as that will illustrate his point.

Inaccuracies based in insufficient data to get a well developed power duration curve.

As a coach if the power duration curve doesn't look right then I look at the data that has gone into it, either not enough or a bad power meter.

Biggest issues, are that people don't challenge the numbers enough or spend too much time testing when they should get cranking on building capacity. This is where the fatigue resistance numbers like TTE give us an indication beyond a power number at FTP, FRC and PMAX.

May not be perfect yet, but it's better than the BS cookie cutter numbers other models give you that don't have a physiological basis. The physiological basis is very individual and very complex!

---

Hamish Ferguson: Cycling Coach

If there's one thing that this thread has taught me it's that all of those things are simply methods of estimating FTP.

I don't suppose you wanted an actual answer, but I'm going to give you one anyway ;) I'm currently estimating my FTP to be somewhere between my 95% MMP20 (244 W) and my somewhat sub-optimal MMP60 effort (234 W). And before you seize upon the opportunity, yes, this is also above my mFTP value, for the litany of reasons discussed in this thread.

No the Troll just wants to Troll.

But your answer again illustrates that people want a number, and generally choose the higher number. Why many like XERT I guess as this always tends to overestimate threshold.

When I assess the demands of the key event I usually find that the rider has the power they need. What they need to develop is the capacity to sustain that power either specifically, a Pursuiter with enough 4min power being able to deliver that power on a specific velodrome at a specific cadence or repeatedly like in a road race with a climb each lap which also means pacing over the race and over the lap.

Hence, early season I would go with mFTP, even if low, it should, in theory, mean a higher TTE number and prompt a larger duration of tempo and sweet spot riding to boost capacity to be able to make the most of the power you have before dialling into more intense efforts as you get closer to the goal event.

---

Hamish Ferguson: Cycling Coach

One shouldn't confuse estimating FTP with the definition of FTP like Alex Simmons in his blog.

http://alex-cycle.blogspot.co.uk/...ven-deadly-sins.html

"FTP is simply be defined as:

"FTP is the highest power that a rider can maintain in a quasi-steady state without fatiguing for approximately 1 hour.

When power exceeds FTP, fatigue will occur much sooner, whereas power just below FTP can be maintained considerably longer".

Apparently approximately 1 hour is anywhere from 30 minutes to 70 minutes, well that is what Coggan says.

Funny Troll, can you do anything beyond cut and paste others ideas?

---

Hamish Ferguson: Cycling Coach