Limitations of TSS

Quoted post said:

Originally Posted by bmoberg337 [IMG]/img/forum/go_quote.gif[/IMG]
... Am I interpreting, or using TSS incorrectly? ...

TSS is an overall workload metric. It serves the same purpose as tracking training hours, miles ridden, kj of work performed or other similar metrics. It has some interesting advantages over some of those metrics in the way it gives you 'extra credit' for bursty intense work but it is still an overall workload metric. It does not tell you anything about the composition of that workload though you can infer some of that by comparing TSS vs hours.

IOW, yes it's possible to rack up a lot of TSS and over time a lot of CTL with high volume low intensity riding and you may feel more tired from less riding that is more intense. So pay attention to ride quality and train to specific needs as a first priority but TSS and CTL still have value to track session loads and long term average training loads. Just think of TSS as a bit more sophisticated version of logging your time or miles on the bike and realize it has the same limitations of just tracking those metrics without also paying attention to how you rode those miles or hours.

-Dave

Dave,

Thanks for the response. Definitely confirms my suspicion regarding the use of the value.

You're not the only one to question TSS and NP. I don't think physiologists yet have a good grasp of how training stress and adaptations interact. It's a tough nut to crack.

...but its not meant to tell you about your adaptions to training - its just a mechanism to track your workload as Dave says. YOU need to work out which composition of that workload will give you the training adaptions you are looking for, and that may, or may not include an overall high TSS score (or CTL etc).

Quoted post said:

Originally Posted by Bigpikle [IMG]/img/forum/go_quote.gif[/IMG]

...but its not meant to tell you about your adaptions to training - its just a mechanism to track your workload as Dave says. YOU need to work out which composition of that workload will give you the training adaptions you are looking for, and that may, or may not include an overall high TSS score (or CTL etc).

My point is, if physiologist's had an accurate model of training stress and adaptation response, we could predict both the type and magnitude of adaptation given the power data as input. Welp, we might also need a whole bunch of other stuff like diet, DNA sequencing, weather, work stress, etc. But let's say we could measure all that too. The problem is that, even if we could measure everything, there is no theoretical model to the plug the data into. Compare this to some other disciplines like physics where there are things like the online bike power calculators.

In any case, I've heard both Dave and Andy Coggan say the same thing about how TSS depends on composition of the training. That statement confuses me a little. Does it mean that you cannot determine which energy systems were stressed during the training and thus cannot predict what type of adaptations might occur? This makes some sense if comparing L2-L5 and L6-L7. If I did 100 TSS of L7 training or 100 TSS of L2 training I would expect different adaptations. But what if I know all my training was L2-L4? In that case, does it make any difference? Should I not expect about the same adaptations from 100 TSS of L2 as compared to 100 TSS of L4? Maybe I don't understand the adaptation chart is Training and Racing with a Power Meter. If TSS does not tell you something about the composition (i.e. intensity) of the training, how is it better than just tracking hours at power level? I guess I thought that was the whole point of TSS.

On a related note, is there any validity to tracking focused L7 work on the same Performance Manager Chart as L2-L4 work? I also wonder if there is any validity to measuring L7 intensity as a ratio of power to FTP. Is there a point at which it makes more sense classify L7 as strength/power training instead of endurance training and measure its intensity as a ratio of force to 1-rep-max force?

I figued out last year that I could artificially inflate my TSS numbers by riding super slow then doing occational all out 1 minute efforts. The overall fatigue of the ride would be low but TSS would be through the roof.

Quoted post said:

Originally Posted by Freddy Merxury [IMG]/img/forum/go_quote.gif[/IMG]

I figued out last year that I could artificially inflate my TSS numbers by riding super slow then doing occational all out 1 minute efforts. The overall fatigue of the ride would be low but TSS would be through the roof.

Yup, I have raised this same issue on the wattage group. I observe it most when I have extended warmups or cooldowns or rides that are split into multiple distinct sections, e.g. 1 hour of L4 intervals following by 1 hour of L2. TSS seems to award too many points to the low intensity sections of the ride. Here is an extreme example for illustration:

FTP of rider is 250 watts.

Ride #1: Ride at 100 watts for 1 hour:

TSS = ((((3600*100**4) / 3600)**0.25) / 250)**2*3600/36 = 16

Ride #2: Ride at 1000 watts for 30 seconds:

TSS = ((((30*1000**4) / 30)**0.25) / 250)**2*30/36 = 13

Ride at 100 watts for 1 hour and then 30 seconds at 1000 watts:

Ride #3: TSS = ((((3600*100**4 + 30*1000**4) / 3630)**0.25) / 250)**2*3605/36 = 147

Even though ride #3 is just the composition of rides #1 and #2, its TSS is almost 10 times as large.

One theory I have about this is that TSS is more accurate for rides where the 10 min (don't know the magic number) rolling average is close to a horizontal line and less accurate for rides where it is not. I think this behavior is an artifact of the way TSS is calculated. It seems to be missing a longer term time dependent weighting (longer term than the 30s rolling average). For example, when it determines the stress the body is feeling at a point in time it weights power inputs that occurred 3 hours previously the same as those that occurred 1 minute previously. Worse, it also includes power inputs that occurred in the _future_. Disregarding the possibility of a significant central governor effect (e.g. semi-conscious upregulating of the metabolism in preparation of work in the future), I don't understand how the body can feel the future.

Another thing I find confusing is Andy Coggan says in velo-fit.comcoggan power.pdfOpen ↗ that normalized power is based off the observation that blood lacate levels are proportional to the power^4, but TSS is actually proportional to power^2 when power is constant.

TSS = IF^2/36

= (NP^2/FTP^2)/36
= (sum(P^4)/n)^0.25)^2/FTP^2/36

When P is constant, this can be simplified.

= (n*P^4/n)^0.25)^2/FTP^2/36
= P^2/FTP^2/36

What am I missing?

BTW, if you do calculate TSS using power^4, you get nonsensical results. The power^2 relationship works much better. I just don't understand how you get from the power^4 lacate relationship to the power^2 TSS relationship.

Quoted post said:

Originally Posted by gudujarlson [IMG]/img/forum/go_quote.gif[/IMG]
What am I missing?

Perhaps you are simply missing the word "may" or "guideline." [IMG]/img/vbsmilies/smilies/smile.gif[/IMG]

Copied from elsewhere, but is typical when reading a lot of these sources......

Quoted post said:

the table below gives some rough guidelines:
<100 low (easy to recover by following day) 100-200 medium (some residual fatigue may be present the next day, but gone by 2nd day) 200-300 high (some residual fatigue may be present even after 2 days) 300 epic (residual fatigue lasting several days likely)

I think Dave mentioned yesterday about taking some of these things way too literal when some of the folk are using words like guidelines, may, likely or other phrases that do not describe absolutes.

I am just glad to have the tools like the power meter and the guidelines developed to give us something that beats just totally guessing, but do not expect those guides to create absolutes or hold those to finite numbers. I know it may be fun stuff to beat these things to a pulp in analyzing so carry on. [IMG]/img/vbsmilies/smilies/smile.gif[/IMG]

Quoted post said:

Originally Posted by gudujarlson [IMG]/img/forum/go_quote.gif[/IMG]

Another thing I find confusing is Andy Coggan says in velo-fit.comcoggan power.pdfOpen ↗ that normalized power is based off the observation that blood lacate levels are proportional to the power^4, but TSS is actually proportional to power^2 when power is constant.

Actually, NP, IF and TSS are all derived from the observed nonlinear relationship between blood lactate and power. The actual observed relationship is p^3.9, but is rounded to p^4 for convenience.

BTW, to those who think they can "trick" NP with alternating short, hard segments followed by recovery segments, I welcome you to try it. It's easy to design a trainer ride with something like 1min on + 1min off. Your theory is that you can do such a ride for an hour at a higher NP than your FTP. I do such rides many times to break up the monotony of constant power rides. I have a CompuTrainer, which makes it easy, but you can also do such rides with a standard trainer and stopwatch. I have found that my max NP for such rides is very close to my max constant power for the same duration.

I think Andy Coggan has said that he doesn't think those recovery guidelines are actually very useful anymore and have been superseded by the PMC.

He also talks about the difference between stress and strain here http://www.cyclingforums.com/t/327227/tss-vs-trimps when comparing TSS to TRIMP. He states that TSS is possibly a better measure of stress (input) and TRIMP is possibly a better measure of strain (effect on the metabolism). For example, TSS does not go up when it is really hot but TRIMP does.

In yet another thread he makes a good point that correlating TSS to perceived post-ride fatigue is not particularly valid. TSS is supposed to be objective whereas perceived post-ride fatigue is subjective. I think that's his point, anyway. It does leave open the question of how do you verify the accuracy of TSS. For example, how can you verify that it is better than tracking hours in zone?

I'm OK with TSS having large error bars and many assumptions and approximations. That's not the point of my last post. I am simply trying to understand the logic Andy Coggan used to derive it, what physioiogical principals it is based on, what assumptions he made, under what conditions is it most accurate and under what conditions is it least accurate. Whenever you use a physical model or a mathematical rule, it's best to understand what the assumptions are so you don't use it in situations where it is not valid and this lead to incorrect conclusions. I would have much more faith in it if I understood it better. That's perhaps just my personality or maybe an artifact of my educational background in the liberal arts.

In the end, I haven't really used TSS other than as a curiosity. I just try to ride 10-16 hours a week during 5 days/week and change up what I'm doing in that time based on my goals. TSS, ATL, CTL, and TSB just flow out of that. I look at the PMC and see it go up when I'm working and down when I'm resting and then I kinda just scratch my head and continue on as I was. I think it might mean more to me after a few years of experience to look back on.

Quoted post said:

Originally Posted by RapDaddyo [IMG]/img/forum/go_quote.gif[/IMG]

Actually, NP, IF and TSS are all derived from the observed nonlinear relationship between blood lactate and power. The actual observed relationship is p^3.9, but is rounded to p^4 for convenience.

BTW, to those who think they can "trick" NP with alternating short, hard segments followed by recovery segments, I welcome you to try it. It's easy to design a trainer ride with something like 1min on + 1min off. Your theory is that you can do such a ride for an hour at a higher NP than your FTP. I do such rides many times to break up the monotony of constant power rides. I have a CompuTrainer, which makes it easy, but you can also do such rides with a standard trainer and stopwatch. I have found that my max NP for such rides is very close to my max constant power for the same duration.

Yup, I agree NP is surprisingly accurate for such rides. It's pretty amazing to me actually. I wish I understood why it is so good. But consider the ride example I gave.

Quoted post said:

Originally Posted by gudujarlson [IMG]/img/forum/go_quote.gif[/IMG]

Yup, I agree NP is surprisingly accurate for such rides. It's pretty amazing to me actually. I wish I understood why it is so good. But consider the ride example I gave.

It all comes back to the observed relationship between blood lactate and power. Blood lactate corresponds with intensity, is consistent with the science of physiological response and is easily measurable. This observed relationship drives NP, IF, TSS and the performance management charts in WKO+. To those who think that NP, IF and TSS are inaccurate, I say, "What do you propose and why?" Frankly, I'm more interested in plowing new ground such as an AWC model as useful and accurate as NP, IF and TSS.

I know Andy Coggan mentions the power^4 blood lactate relationship. I linked the article in which he states that. The issue is that TSS is proportional to power^2 when power is constant; not power^4 as one might expect. There in lies one of my confusions.

One thing I forgot to point out in my example is that I excluded the 30 sec rolling average in order to make the calculation easier. The actual TSS for the ride is going to be lower than 147, but it will still be much higher than the sum of the TSS from the 2 other rides. Think of my example qualitatively not strictly quantitatively. I'm not trying to create a "NP buster" as some have tried. Rather I'm trying to illustrate a (surprising) behavior. The behavior I'm trying to illustrate is that events separated by a great deal of time (e.g. 1 hour) have a great deal of coupling; more than I would expect.

Quoted post said:

Originally Posted by RapDaddyo [IMG]/img/forum/go_quote.gif[/IMG]

BTW, to those who think they can "trick" NP with alternating short, hard segments followed by recovery segments, I welcome you to try it. It's easy to design a trainer ride with something like 1min on + 1min off. Your theory is that you can do such a ride for an hour at a higher NP than your FTP. I do such rides many times to break up the monotony of constant power rides. I have a CompuTrainer, which makes it easy, but you can also do such rides with a standard trainer and stopwatch. I have found that my max NP for such rides is very close to my max constant power for the same duration.

That's not my experience. I've done similar rides as "Freddy's" (30" all out with ~5' of easy riding in between the efforts) and have 45'-1 hr NP's that are much higher than I could achieve with a much more modest VI.

We're talking about "NP Busters" - right?

Quoted post said:

Originally Posted by dkrenik [IMG]/img/forum/go_quote.gif[/IMG]

That's not my experience. I've done similar rides as "Freddy's" (30" all out with ~5' of easy riding in between the efforts) and have 45'-1 hr NP's that are much higher than I could achieve with a much more modest VI.

We're talking about "NP Busters" - right?

My sprint workouts are generally more than 100 TSS/hour, but not much more; maybe 5% at most. I don't expect much more accuracy than that. Anything in sport science with 5% error is doing pretty awesome. I think the bigger issue is with rides that are made up of distinct sections of different types of efforts where you can observe what seems to be (at least to me) errors on the order of 500%. It's harder to estimate the error because there is no neat rule for lower intensity rides like the 100 TSS/hour limit for intense rides.

Quoted post said:

Originally Posted by gudujarlson [IMG]/img/forum/go_quote.gif[/IMG]

I know Andy Coggan mentions the power^4 blood lactate relationship. I linked the article in which he states that. The issue is that TSS is proportional to power^2 when power is constant; not power^4 as one might expect. There in lies one of my confusions.

One thing I forgot to point out in my example is that I excluded the 30 sec rolling average in order to make the calculation easier. The actual TSS for the ride is going to be lower than 147, but it will still be much higher than the sum of the TSS from the 2 other rides. Think of my example qualitatively not strictly quantitatively. I'm not trying to create a "NP buster" as some have tried. Rather I'm trying to illustrate a (surprising) behavior. The behavior I'm trying to illustrate is that events separated by a great deal of time (e.g. 1 hour) have a great deal of coupling; more than I would expect.

30 second smoothing window of Normalized Power is weighted to ^4 to reflect the exponential rise of blood lactate in relation to intensity.

"This algorithm is somewhat complicated, but importantly it incorporates two key pieces of information: 1) the physiological responses to rapid changes in exercise intensity are not instantaneous, but follow a predictable time course, and 2) many critical physiological responses (e.g., glycogen utilization, lactate production, stress hormone levels) are curvilinearly, rather than linearly, related to exercise intensity"

TSS is then calculated by raising IF to ^2.

I don't think I'm tricking anything. I know that if I go ride for an hour at 120w and do three one minute intervals at 650w. I will end up with a normalized power number much higher then what I could actually average for an hour. When my IF for an hour ride is 1.3 and my average power is 175w my TSS reflects the IF to give me a number which I don't believe accurately reflects the physiological effect of the ride.

Quoted post said:

Originally Posted by frost [IMG]/img/forum/go_quote.gif[/IMG]

30 second smoothing window of Normalized Power is weighted to ^4 to reflect the exponential rise of blood lactate in relation to intensity.

"This algorithm is somewhat complicated, but importantly it incorporates two key pieces of information: 1) the physiological responses to rapid changes in exercise intensity are not instantaneous, but follow a predictable time course, and 2) many critical physiological responses (e.g., glycogen utilization, lactate production, stress hormone levels) are curvilinearly, rather than linearly, related to exercise intensity"

TSS is then calculated by raising IF to ^2.

Right, so if you work out the math; see above, NP is proportional P^1 when P is constant and TSS is proportional to P^2 when P is constant. There's no x^4 relationship between either NP or TSS and power. Note that I assume that the lactate data Andy Coggan used came from steady state (i.e. constant power) efforts. Obviously, there is something I don't understand about the derivation of NP and TSS starting from the lactate data fit.