Dylan Johnson's "The Problem with TrainerRoad Training Plans": it's gonna be a busy day around here

[Joelrivera]

Thats usually how a good run forum do.
As long as the thread doesnt cross certain lines, they will more than likely allow it…

[timon]

@Captain_Doughnutman

Yup, I guess so. All that does is distort the analysis of the stand-alone TR plans thus making all that “big data” kinda useless.

I don’t necessarily think so. I don’t work in data at all, but I’m guessing they can pretty easily run queries that would show for instance …male users, 3.5w/kg or greater, age 25-34 who average 10-15 hours per week on a bicycle. then divide that group into those who trained x% at z1/z2 vs those that trained y% of the time at z1/z2 and then see if that had any statistical significant vs their top 10min power over a 2 year period. Could also make it a function of users who spent some percent of their time inside/outside or on a strict TR plan or not.

I’m sure that specific example has issues, but a decent sports scientist would be able to run enough queries and sift through the data and weed out the flaws and draw some solid theories/conclusions.

[tuco34]

Disagree. DJ has a valid point.

[CoryE]

I hope no one strikes a match, there sure are a lot of straw men up in here.

[Captain_Doughnutman]

@timon

I’m guessing they can pretty easily run queries

How will TR extrapolate the efficacy of a straight TR plan if it’s been altered?

:

Addendum:
How will TR extrapolate the efficacy of a non-TR plan vs a TR plan?

[patrickhill]

Just finished my 541st Trainer Road ride in less than. Such a great platform and tool. I’m going to throw out some thoughts:

• The plans work better for someone with a rapidly rising FTP. I tacked on 114 watts in two years. Athletes who are relatively new to structured training can handle all of the intensity because their FTP has already blown past their assessment by the second week and they are actually operating at a lower IF than prescribed.
• Once you have plateaued, keeping up with the intensity becomes almost impossible, especially if you know how to dig deep on a ramp test. It may take a few years for this to happen, and much of the user base hasn’t gotten there yet. But when it happens, especially if you’re a bit older, it isn’t pretty. And if you miraculously gut it out and make it through (like I did with SSBHV1), you’ll probably overtrain and see a fitness drop like I did.
• Dylan’s assessment was eye opening for me. I was thinking, holy crap, maybe it’s not me after all? I cringe every time the podcasters tell me I just need to “get used to the discomfort” … of doing 4-5 highly demanding and intense workouts per week (with an FTP that’s also calculated too high, btw).
• On that note, what Dylan doesn’t mention is that the FTP assessment protocol doesn’t work for many many users. There is nothing that takes a users specific power curve into account. I can handle the hardest sweet spot and anaerobic rides based on the ramp test, but I need to reduce the intensities of longer VO2 and threshold workouts by 5% in some cases (that’s a lot). You have to somehow figure out on your own how you need to adjust each type of workout, and this can take a long time to get right. Some users don’t have that kind of patience. As it is now, it goes from being to a nice objective way to setting ftp to a subjective mess pretty quick.
• I am discouraged that I haven’t seen the plans change much if at all in the past 3 years, despite tons of new data rolling in, BUT I am fully confident that @Nate_Pearson, @chad, and the TR team will solve these challenges and deliver an even better product. I will happily pay for advanced features.
• The podcast is AWESOME
• The podcast is much better when @Nate_Pearson is there. He cracks me up, is a great example of someone who can achieve big gains with the platform, and he and I are always yo-yoing as to who’s FTP is currently higher.
• The podcast is much better when @chad, is there. This guy is a genius and gives it to you straight, but I wonder if a bunch of users with rapidly rising FTPs are skewing the TR data he is getting.
• I am now using 72% of the highest 1 minute ramp test power to set my FTP (rather than 75%)
• I have switched everything to low volume so that I don’t have any more than 3 days of intensity per week. It is interesting to note that @Nate_Pearson pretty much said he is doing the same thing a few weeks back.
• I am adding as much endurance and recovery riding as I can during the other days. I am confident that given the same TSS, I will feel better and be faster this way vs. using the mid or high volume plans, and I would not be surprised if the standard or potential future masters plans look more like this.
• Let’s not let Dylan’s snark get in the way of making TR even better. I am looking forward to it.

[chad]

@batwood14

[chidlow]

@timon

I don’t work in data at all, but I’m guessing they can pretty easily run queries that would show for instance

I do work in data, and no it isn’t that simple. I had a post that deconstructed why, in response to one of Nate’s forum posts, but that thread recently got deleted and I’m not gonna type it all out again. But basically - sample biases. Self-selection bias and survivorship bias in particular.

Edit: because @mcneese.chad is a legend he dug up my post for me without me even needing to ask

I’d like to make a general point here, as someone who has either been working as a data scientist or been heavily involved in that space for many years. I have no idea of how TrainerRoad uses their data and thus no idea if what I’m about to say applies in this situation, so hopefully this is taken in the philosophical sense it is intended as, rather than as a debate about any specifics. Although if TR is open to sharing more details about their analyses, that would be pretty cool.

It feel like there is a modern trend of holding “the data” up on a pedestal, as if data is some naturally occurring phenomena that, thanks to the power of computing, we’re merely going out and harvesting en masse for the first time in human history. Unfortunately, you cannot separate data from the context in which is was collected/created. Equating data with facts is a common mistake, one that I’ve seen even experienced data scientists make.

Most of the problems which aren’t something simple like measurement error, boil down to some form of sample bias. And when the data is used in a way that influences further data collection, your problems can compound and you end up in a sort of “feedback loop”.

Here’s a classic example of how data can be misleading if you don’t take into account the potential biases (in this case survivorship bias). To be more concrete and relevant to the discussion (and again, I have no idea of the statistical modelling that TR is doing so I’d have no idea if they’re accounting for this - I’m just using this as an example), there is almost certainly an element of survivorship bias in the TR dataset - if either the Ramp Test or the TR plans don’t work for you, you’re unlikely to continue your subscription.

Or maybe they work for a while, and then you plateau because the newbie gains are over, and you start looking at alternatives, maybe not following TR plans exactly, you make some gains here and there, and then maybe you end your subscription. If you just look at the data, that sort of pattern would seem to suggest that people who follow TR plans make the biggest gains. Whereas in this example (and again, because I’m not looking to weigh in on any debate about either the ramp test or TR plans, this is just a hypothetical example), it was potentially more influenced by the fact that they were untrained.

There’s surely other biases in the data too - I doubt that population sample of TR users could be considered an entirely random of all those who ride bikes and train.

So I guess what I’m trying to say is, whilst I’m sure TR’s dataset is invaluable, and could be a source for further insights into training methods, I don’t believe that its size alone justifies considering it above physiological understanding. I’m currently working with a dataset that generates billions of data points per day, and thankfully we actually have a way of estimating and correcting for the main source of sample bias to generalise insights to the “universal” context in which they occur, and yet we still have to be very careful to consider the context in which the data is gathered - or else it’s all too easy to be misled.

[Captain_Doughnutman]

@chidlow

Self-selection bias and survivorship bias in particular.

This is what I was inelegantly getting at — eventually the TR user base will be mostly people for whom TR plans work.

[Joelrivera]

And just like that we should ALL know that this has been noticed by the administration!

[Captain_Doughnutman]

I’m only here to bag #1000.

[trainingRan]

Just my two cents.

I think the key point of training is to do training with consistency.
It’s the secrete to cycling and also anything that requires dedicated
effort to improve. It sounds easy but it’s indeed really hard.

TR makes It Obvious

It provides clearance. The user has a clear goal of what to do today, this
week, and this month. This reduced mental stress will ease some
hardship of the intense training.

TR makes It Attractive

I always upload my training data to Strava and I believe a consistency
update of my training data will make me an attractive cyclist.

TR makes It Easy

TR makes it really easy. I just jump into my bike and start a training
session. I will make the preparation done before the training session
started to reduce any friction before I start to train myself.

TR makes It Satisfying

See a consistent calendar is a reworking process. It makes me feel
satisfied.

[chidlow]

Yeah to which Nate replied something along the lines of “yeah we’d considered that but we also have Strava data”. Which I didn’t bother responding to because honestly it just sounds like wishful thinking, because the problems still stand:

1. they self-selected into TrainerRoad to start with
2. they self-selected into Strava import
3. unless something shady is going on, TR shouldn’t continue to receive Strava data after people leave.

[tshortt]

@tshortt

ay what you want about self-coaching (I hated it and was crappy at it), but no coach is going to let a bunch of know practically nothing amateurs pitch a fit over what they’re doing or not doing. You’re more likely to hear: “go play for the other team, slick”.

@batwood14 You know what I’m talking about. Good on you.

[ajsds66]

Having a demanding job, a family and trying to read all the posts in record time for the past 24 hours has kept me in zone 5 most of the time. I hope TR knows how detrimental this is for me and will tailor a training plane accordingly.

[TRMailloux]

Only one way to settle this, Nate vs DJ heads up at Cape Epic.

[chidlow]

I’m definitely putting this into TrainingPeaks as “Other workout”. How many TSS to give it though

Me catching up with this thread from overnight:

[rbynon]

SST or POL? There is a happy medium that DJ touched upon, and one I use (without really knowing I did) is the pyramidal approach. A little HIIT, maybe a bit more zone3/4, the rest endurance, but not a crap ton of it like in POL training. The devil is in the details, though, and I’m not aware of any off-the-shelf plan, TR included, that offers one.

[Captain_Doughnutman]

A person could get relatively fast on 6hrs/wk:
1x4hr Z2
1x1hr VO2max: 5x4min
1x1hr Threshold: 2x20min O/U

[palmerneill]

You support this characterization: " allowing people to congregate in their yard (this forum) to take a shit on their house (their product) without shutting it down."

I hope not, because otherwise I just lost a lot of respect for you. If anything that comment should have been censored, not applauded.