If your workouts are scheduled on a 7 days basis then yes. It is really looking at the period of your training where the pattern of workout types repeats. Then looking across 12 of those such periods. If you are a masters athlete and your pattern runs over 9 days, then that would be a better period to compare.
Plus it will look different if the weeks include events, or a transition from base to build to speciality etc. It will look different for someone doing progressive overload through hours compared to someone doing it by load.
I ride ultra distance and in an event week can match the hours that would normally be 2 months in training. Context is everything to understand the numbers.
Yeah, I did the same but it’s quite close (13.2 instead of 13.8), so doesn’t change what I already knew - that I just ride a lot without any structured training!
I’m also curious if weekly TSS for 12 weeks would make more sense. I had a big 22 hour week (bike packing trip), but it was only slightly more TSS than the next week of 16h - from a load perspective the big hour week wasn’t overly taxing compared to what the higher hours might imply.
Just ran the numbers for TSS assuming a 5% ramp rate for progressive overload, and a 20% drop for a deload week every fourth week. Then continue progression. If you follow that perfectly then the coefficient of variation is 15.1% using sample stdev. This is independent of your weekly TSS as both mean and stdev scale proportionally.
I’d agree that they should have the same mean, but the standard deviation could be wildly different.
As an extreme example of if I rode 10 hours yesterday(Sunday) then 10 again today(Monday), it would show up on a platform like TR as 10 hours for week 1 and 2. We know that’s a very different training stimulus than doing 5 two hour rides throughout the week.
I know there are a lot of people who are bouncing between missing and extending workouts. I’d bet there are a lot of people who think they have it under control because they are getting close on the simple weekly number they monitor. They are mixing in Sunday long rides and days off watching football based on what happened the last 6 days of training, and aren’t really taking it into effect for the weekday training after Monday’s reset.
But whether you set the start of the 7 day period as a Monday or a Wednesday you still come up with 20 hours per week. Which was your original question. Does the start day of the week / period make a difference to CV?
Post up an example 12 week period and your workings of the numbers. Happy to be proved wrong, if it’s not the case.
Ok, I ran it again through AI using TSS (sorry for the AI material..):
Result
Average weekly TSS:899
Standard deviation:67.5
Coefficient of variation:7.5%
Comparison with Hours
Metric
CV
Weekly Hours
13.8%
Weekly TSS
7.5%
Your TSS is much more consistent than your hours. This suggests you’re maintaining a fairly stable training load despite fluctuations in volume. The 22.5-hour week was a large outlier in terms of hours, but it did not produce a similarly large jump in TSS, indicating that week was likely lower intensity (more endurance riding) rather than a huge increase in overall stress.
I do think using hours ridden in a day doesn’t seem like the best metric to base tis on. A 1.5 hour sweet spot workout isn’t the same effort as a 1.5 hour endurance workout. So alternating a week of three 1.5 hour sweet spot workouts followed by one with endurance workouts isn’t consistent week by week.
I guess in general hours works as i highly correlates with effort a person puts out but isn’t that good.
Seems like a more daily type of metric like CTL/ATL/TSB type graphs would be more useful. We would want a line on the chart that represents the variability of the ATL line. Take the last 42 days of ATL (42 comes from the same window size as CTL) and compute the (stdev/average)*100 and plot it as VTL (variable training load, I’m not good at names so…) ATL is based on exponentially weighted moving average so this may result in a smaller number then the consistency score metric but should still show the variability
I feel like this avoids the need to care about how long your training week is and the 12 weeks look seems arbitrary vs the reason CTL picked 42 days (not saying 42 is right, but could at least use the same window size of CTL and VTL)
Unpopular opinion: cyclists don’t need another metric, one glance at the Performance Management Chart is enough to show if you have been consistent.
But, If you really want to nerd out, I would plot a 12-week rolling z-score or such, which would give, not only a “consistency” number, but also show how and how far you deviated from the mean.
z-score is just how many standard deviations you are from the mean. If you have a large variability your standard deviation would be larger so you could have a low z-score and yet still be highly variable.
Not an unpopular opinion. In fact, it’s a rule. So if you are going to use this metric you must choose another that you are currently using and foresake it going forward.
I was looking at making the CV calculation and was surprised at the numbers from my plan throughout July, which was a mess with travel & weather. It looks like a winter trainer block, despite having a lot of days off the bike.
Week 8 - June 29
7:11:00
Week 9
9:43:00
Week 10
7:45:00
Week 11
9:43:00
Week 12
7:39:00
I’d also want to dig into where the 12 hr over 4 days “training camp” hid in those numbers. If I changed week 8 to start on June 27th instead of the 29th, and analyze a Saturday-Friday 7 day period I get the following weeks.
Alt 8 - June 27
9:52:00
Alt 9
7:42:00
Alt 10
11:53:00
Alt 11
5:46:00
Alt 12
8:41:00
It gives a different story and calculation entirely with ~2x standard deviation. Most importantly it shows that cram training and time off more clearly, which would be important to call out when analyzing training history.
Now if we go for a 40% reduction in load during a deload week every fourth week. Plus start with a weekly TSS of 300.
For two 12-week TSS progressions:
Progression
Mean weekly TSS
Standard deviation
CV
8% build / 40% deload
348
76
21.9%
5% build / 40% deload
317
62
19.7%
The coefficient of variation is relatively high because the deload weeks create large drops in TSS. That’s expected and what you want: you have progressive loading but also clear recovery weeks.
If you calculated coefficient of variation only across the build weeks, it would be much lower.
TR told me how to calculate the metric, but when TR presented four athletes & representative scores, the method TR gave does not produce the metric scores they presented. TR are taking a 13 week average & not a 12 week average…and I have yet to figure out how they are coming up with the scores.
But, I guess I should just shut my stoopit mouth, because nobody else seems bothered by it…or even notices even after it’s been pointed out. But, anyhow, TR says, ‘Here is an actual athlete with a score of 23.’ I say, no, not if you use the metric you actually described. Then, you know, TR drew some pretty insightful conclusions based on that metric.
But, I have to ask myself if those conclusions would still stand were TR to use the metric they actually prescribed instead of whatever methodology they used to generate that slide.
Either way, this thread has been delightful. Probably if athletes just walk away with the thought that they are going to improve their consistency, then the forum has been well served.
Yep, I didn’t need to do the calculation to tell me that, but was interested in the number regardless. I don’t follow any progressive overload or even do intervals - I’m fully aware my fitness could be higher, but I really just like riding during the warm months. I’m more disciplined with structured training from Dec-Mar, but that’s pretty much it.
Anyway, I’m a good example where the number is too low from optimal.
That’s what I’d want to know, and have a tool that lets me track over time. Probably could set that up with more time from a data export, but I used July because I had already filtered the rides out by day. However I quickly see another set of weeks that turn into 16 & 7 hour outliers.
I have a lot of variation in Sunday rides, and I think a lot of it is because a basic “rule”’ is aiming for the week-week consistency which is readily available on every platform going into Sunday. Monday-Saturday has X hours so I need Y to be in the weekly range. Instead of sticking to the 3-4hr planned, there are a lot of 1-6 hr rides. Inconsistency introduced in order to make the measured consistency number look better.
But, anyhow, TR says, ‘Here is an actual athlete with a score of 23.’ I say, no, not if you use the metric you actually described. Then, you know, TR drew some pretty insightful conclusions based on that metric.