Nail on the head. So many things impact on human performance that trying to predict it based on only one measure, power output in isolation, will always be more art than science.

That said, those who strive to improve the science should be encouraged.

Nothing secret about it, as I've been talking about the relevant issues* for years and years. Most people apparently just haven't been paying close enough attention to put two plus two together. Then I do a series of webinars in which I lay out my ideas in a bit more detail, and people fall all over rushing down the same path while pretending their thoughts and ideas are completely novel.

*E.g., "it is called training stress score and not training performance score for a reason", "the impulse response model tells you when to train, not how/how much to train", etc., etc., etc.

That's not it. Roady objects to the smoothing/weighting/averaging/unweighting approach that I pioneered in coming up with TSS since it can be expressed as a theoretical power, either not realizing or overlooking the fact that that was just a side effect of the real goal, which was to predict the overall physiological strain resulting from a workout based on the intensity and the duration.

IOW, while the approach has proved accurate enough for it to have other uses - e.g., for estimation of FTP, as a global constraint in developing pacing strategies (e.g., as implemented by, although not original to, Best Bike Splits), for predicting whether a novel interval workout is "doable" or not - was simply fortuitous.

Nonetheless, many people find it just fine in those ways. I don't use it to tell me when to train, because I generally train every day, but I do use it to inform my decision about how much to train each day. And it seems to work fine as a performance score / adaptation score, indeed it puzzles me why a study would be unable to detect a relationship like the one below, which shows my performance vs CTL since resuming training last November after 3 months off:

I was expecting that, once the road race data was removed, leaving the time trial data, the study would detect a relationship.

It works for you as you have illustrated, perhaps not for others?

Perhaps the problem is due to the data being collected outdoors, rather than indoors?

Also your training is very highly structured and tailored to time trials. Perhaps those who are training for other events and or outdoors, cause more noise in the data?

Here is another really big hint, in a graph I posted to FB the other day...if somebody were smart enough, they could reverse-engineer the WKO4 model from this:

This is a niggling detail, but that's not actually the impulse-response model, or even a watered-down version thereof (i.e., the Performance Manager). Instead, you've plotted FTP against the equivalent of the "fitness" term, leaving out the "fatigue" (or "freshness") aspect.

Anyway, yes, if you keep training composition relatively constant then CTL provides a measure of adaptation as well as strain...it just wasn't designed, nor is it ideal, for that purpose.

Yes, I've yet to find a way to incorporate TSB into an improved numerical predictor of performance compared to just using CTL. I've even devised my own TSB2 to address one of the main issues I've found with TSB, but while it helps in terms of the general trend, I've yet to come up with something that can accurately predict day to day variations in performance. CTL seems to be a good predictor of my performance provided I'm not "excessively fatigued", but a metric to accurately predict whether or not I am excessively fatigued on any given day has eluded me so far. One fairly major issue is that TSS doesn't work well for me as a stress score, i.e. whether or not I am excessively fatigued on a particular day depends a lot more strongly on what I've done in recent days than on how much (as measured by TSS) I've done in recent days.

I didn't read the study.... but i wonder if the relationship is broken because numerical models focused on physical performance may struggle to quantify psychological factors. For example, I'm pretty deep into a IM build and while my TSB isn't that low overall, because my overall CTL is high, and because I ramped up so much higher in my training thsi year vs. last year, I'm up against phycological limits (burnout) and general overreaching. It's not a bad thing, as I'm testing my limits. But how I feel isn't perfectly represented in the model. IF I were to race today, I'd perform far worse than the PMC would predict. Btu a couple more light training days and I'll be OK.

I think this become part of the art of coaching where science can struggle. It also cannot measure other outside inputs like stress at home, work, family, time spent standing vs. sitting, volume of sleep, quality of sleep. It makes some assumptions on that , or you need to adjust some of the constants.


That brings an intesting point. I wonder if the time constants need to change for recovery rate as a CTL is sustained after increasing rapidly early in a season and held at a certain level. I.e... peaking too early, or not resting enough after an early season peak then building back up. The model assumes you can sustain a training load indefinitely.

---

TrainingBible Coaching
http://www.trainingbible.com

In part, that could be because you're focused on TT power, whereas it is short-term power that it really impacted by how "fresh" you are. For example, there's a study soon to be published in Med Sci Sports Exerc showing that even doing intervals the morning before followed by 2 h of tempo the afternoon before only slowed cyclists by 1 min in a simulated 40 km TT compared to when they were fully rested. When you consider studies like this one (or frequency distributions of TSB at time of a power PB for shorter vs. longer durations), it really becomes rather worrisome that the impulse-response model (upon which the PMC is based, and as fully implemented in RaceDay) has never been validated using a performance test longer than 5 min in duration. But, these are probably more thoughts that are going to be lost on 99.99% of people...

You seem to have missed a key point re. the PMC. While Banister's model attempts to quantitatively predict performance, the PMC requires that you balance the notions of "fitness" and "freshness" more abstractly to determine what timing AND amount of training yields the best performance (note that the classic impulse-response model ignores the amount aspect, simply assuming that more is always more). IOW, you should be looking at your CTL as well as your TSB, and based on what you've written it sounds as if the PMC approach is working exactly as it should (i.e., you've found your personal CTL limit, or something close to it).

So then if, for a given season you've determined your CTL limit, then you can put together training loads to reach that limit at the ideal time in the season to taper from it for a "A" race. Loading beyond that point would simply risk injury and burnout. Obviously it's somewhat of a moving target.

I do like the ability to tweak training load here and there and see the impact for a "B" race where I want to be at maybe a 0 to +5 TSB for each discipline, and then look at how much fitness I give up in a taper for an "A" race and adjust training loads the 3 weeks prior to move that number around a little. Then compare my numbers to my previous "A" race where I performed well and ideal, try and have a slightly higher CTL and about the same TSB going into the next one.

One question... and this might be in the Training Bible, but the longer the distance, do you want a comparatively higher or lower TSB than for a shorter distance "A" race.... given a similar overall training load for the season? OR about the same.

What I've learned most from the PMC model, is that since you have a personal limit on what CTL you can reach and maintain, the impact of time off or an injury is not as big as you might expect with a reasonable time from to train following recovery. Where I'm at, if I drop my CTL in 1/2, I can recover that fitness it in about 2 build cycles it seems and still have good form with 2 weeks more for a taper. SO 10 weeks out, from an "A" race and an injury is NBD. Closer than that and you may struggle. It makes you reconsider how aggressively your train following the "Base" period... with running in particular.

---

TrainingBible Coaching
http://www.trainingbible.com

What was Chris Froome's TSB and CTL going into today's stage?

Sorry, you won't find the answer there. ;)


In general, the shorter the event, the higher TSB should be. For a more detailed answer, please see pp. 155-158 of the 2nd edition of our book, or (for a more tri-specific answer), Dr. Steve McGregor's chapter in this book:

http://www.amazon.com/...-Friel/dp/1450423809

I don't know (but I could find out).

If he was using a Stages power meter the data would be dubious.

Can you also find out what pain killers he was on and what drugs he was taking for his asthma?

My theory is that only person on the planet is smart enough to put a power meter to good use. And to emulate the researchers in this profession, no attempt will be made to reject the null hypothesis.

I know TRIMP has been validated. Is it the case for TSS? If so, do you have a reference? Thanks.

TSS has never been scientifically validated.

Define "validated".

Is TSS measuring what it is supposed to measure? All measures need to be validated (using whichever adequate concept
of validity, be it convergent, construct etc.) just to make sure they do where they're supposed to.
See http://www.ncbi.nlm.nih.gov/pubmed/21904234 for TRIMP for instance. I'm not implying in anyway that TSS is not valid.
I was just looking for TSS validity and haven't been able to find anything thus far (OK, I haven't spent a lot of time searching...)

Andrew Coggan has been asked this question before.


Question; Dr Coggan's TSS was inspired and modeled on Dr Eric Bannister's heart rate based training impulse (TRIMP) which has been validated by numerous scientific studies.

Has TSS been validated in any scientific studies?

Andrew Coggan's reply;

"No (even though I've been encouraging somebody to take up the bit for years). "

thanks

http://thetriathlonbook.blogspot.co.uk/...elated-rambling.html
""
Training Stress Score (TSS), is a training load estimator for cycling invented by Dr. Andrew Coggan, and is modeled after Dr. Eric Bannister's heart rate-based training impulse (TRIMPS). It takes into account both the intensity (i.e., IF) and the duration of each training session, and according to the author, it “might be best viewed as a predictor of the amount of glycogen utilized in each workout.”

TSS has the advantage that it is easy to calculate and that it is based on the direct measurement of the applied stimulus (power), unlike TRIMPS that is based on heart rate. However, unlike TRIMPS, TSS has not been validated in any scientific studies, which means that its use by many comes from believing that it is a good tool to estimate training load.""

The problem is so many people believe TSS has been scientifically validated and no longer look at heart rate. In my opinion you should look at power and heart rate.

To quote Frank Day: ugh. What a crappy study - [of course two different measures of training load calculated from the same (heart rate) data will be correlated with each other. The only way you wouldn't see such auto-correlation would be if the data were manipulated in radically different ways.

If you're looking for a study validating TRIMP (as a predictor of physiological strain), these are probably as good as it gets:

Busso T, Hakkinen K, Pakarinen A, et al. A systems model of training responses and its relationship to hormonal responses in elite weight-lifters. Eur J Appl Physiol 1990; 61: 48-54.

Busso T, Hakkinen K, Pakarinen A, et al. Hormonal adaptations and modelled responses in elite weightlifters during 6 weeks of training. Eur J Appl Physiol 1992; 64: 381-386.

Even then, though, you can't really separate the validity (value?) of TRIMP as a predictor of physiological strain from the impulse-response model itself, since they didn't look directly at the relationship of TRIMP to hormonal response (e.g., during/after a single bout of exercise).

Anyway, back to TSS: when I first proposed it back in 2003, it was the first ever objective, stress (i.e., input)-based measure of training load, with its purpose being to serve as input function when modeling the relationship between training and performance. Despite regular encouragement from me, the scientific community has unfortunately been quite slow to get around to studying the idea (there are reasons for that, but no time now to explain). A decade or so on, though, things are starting to change. Specifically, in addition to Fergie's quite-commendable effort several other abstracts/papers have utilized/assessed TSS and/or one of its progeny/imitators/components (with Phil Skiba leading the way back in 2007):

Skiba PF. Evaluation of a Novel Training Metric in Trained Cyclists. Med Sci Sports Exerc 2007; 39: S448.

(See below.)

http://www.ncbi.nlm.nih.gov/pubmed/19910822

(Demonstrates that rTSS/the PMC can be used to predict running performance.)

http://www.ncbi.nlm.nih.gov/pubmed/20058020

(Demonstrates that variations in TSS/CTL predict variations in Hb mass.)

http://www.ncbi.nlm.nih.gov/pubmed/21113616

(Used TSS and IF to match/compare the training of two groups of cyclists)

http://www.ncbi.nlm.nih.gov/pubmed/24405984

(Demonstrates that the running equivalent of normalized power is a predictor of optimal pacing strategy.)

http://www.ncbi.nlm.nih.gov/pubmed/24104194

(Demonstrates that the running equivalent of TSS is a better predictor of training-induced improvements in performance than either TRIMP or Foster's session RPE.)

Of the above, Phil's original study is probably most directly on-point, so it is probably worth reproducing the abstract here (with some emphasis added):

"Numerous systems have been developed to quantify athlete training, many based upon subjective criteria. Recently, a novel system based upon lactate-normalized power output has been popularized for cycling (Coggan 2003, 2006), which has not been evaluated in the literature. This system should be superior to existing methodology because it relates a purely objective parameter (power output) to resultant metabolic stress by weighting cyclist power output with a 4th power function that closely tracks serum lactate response to a standard ramp exercise protocol. This value is then compared to the average power an athlete is capable of maintaining for one hour (previously shown by Coyle et al (1988) to be highly correlated to power output at LT) to generate a training stress score.PURPOSE: This investigation examines the validity of this algorithm in a group of trained cyclists (n=5). This work also evaluates the utility of the related training stress scoring system in the quantification of training load and performance modeling using convolution integrals.METHODS: Power meter files for one-hour (range 51–62 minutes) individual time trial races (ITT) and one-hour (range 51–60 minutes) criterium races (CRIT) were obtained from 5 trained cyclists. Average power (AP) values were compared between ITT and CRIT. CRIT power data were then subjected to a 4th power-weighted 30-second moving average to generate a normalized power (NP) value. ITT AP and CRIT NP were then compared. Training stress scores were generated and used as the input function for systems-based performance modeling for a national-level track cyclist per the method of Morton et al (1991).RESULTS: CRIT AP was highly correlated with ITTAP (p<0.04, r2=0.791), however, CRIT NP was more highly correlated to ITT AP (p<0.001, r2=.978). Using the examined training stress score, it was also possible to accurately model performance (p<0.0001, r2=0.9189). CONCLUSIONS: Though additional work with a larger sample size is required, these data indicate that NP may be superior to AP in describing how strainful a variable-power work task is. These data also demonstrate the utility of the associated training stress quantification system in performance modeling for trained cyclists."

In addition to the above, a number of other peer-reviewed studies have also used, or at least cited, some of my other ideas, e.g.:

Abiss CR, Quod MJ, Martin, Netto KJ, Nosaka K, Lee H, Suriano R, Bishop D, Laursen PB. Dynamic pacing strategies during the cycle phase of an Ironman triathlon. Med Sci Sports Exerc 2006; 38:726-734.

Gregory CM, Doherty AR, Smeaton AF, Warrington GD. Correlating multimodal physical sensor information biological analysis in ultra endurance cycling. Sensors 2010; 10:7216-7235.

Francis JT Jr, Quinn TJ, Amann M, Laroche DP. Defining intensity domains from the end power of a 3-min all-out cycling test. Med Sci Sports Exerc 2010; 42:1769-1775.

Robinson ME, Plasschaert J, Kisaalita NR. Effects of high intensity training by heart rate or power in recreational cyclists. J Sports Sci Med 2011; 10:498-501.

Cowell JF, McGuigan MR, Cronin JB. Movement and skill analysis of Supercross BMX. J Strength Cond Res Publish Ahead of Print 2012 (DOI: 10.1519/JSC.0b013e318234eb22)

Note that there may be others out there, since as merely a hobbyist in this arena I don't make a practice tracking each and every citation of my work...