FTP Test - PowerTap vs CompuTrainer Comparison

Porkyboy said:

Hi

In case anyone is interested I've done my latest, and almost last, PowerTap vs CompuTrainer comparison on data collected during an FTP test I did last night according to the recipe in "Training and Racing with a Power Meter". The findings are on my blog, link in signature file. Will try to do last test (a simple final step up test) tomorrow.

Regards to all.

PBUK

You stud, you.

Over 5 hours over 350, 16 hours at a more leisurely pace with the middle 15 hours at over 300 watts. Wow...

... and you sounded like you were worrying about La Marmotte! Humbug... 39x23 @ 80rpm on the Alpe for you lad!

Andy wilkinson's End to End record would be in danger if you rode it - unless of course the bottom scale should have read seconds and not Minutes.... 😉

Hi

swampy1970 said:

You stud, you.

Over 5 hours over 350, 16 hours at a more leisurely pace with the middle 15 hours at over 300 watts. Wow...

... and you sounded like you were worrying about La Marmotte! Humbug... 39x23 @ 80rpm on the Alpe for you lad!

Andy wilkinson's End to End record would be in danger if you rode it - unless of course the bottom scale should have read seconds and not Minutes.... 😉

Looks Ok to me! :o

PBUK

Porkyboy said:

In case anyone is interested I've done my latest, and almost last, PowerTap vs CompuTrainer comparison ...

Just a small point about comparisons. I agree that you can tell just by looking at the plot that they are extremely close to each other and quite frankly, I think that you could just leave it at that; it's close enough for non-professional stuff.

Since you seemed to want to crank out some analysis though, I'd say it's helpful to compare the averages over certain intervals as you have done but overall it's better to compare the difference between the two on a point-by-point basis, i.e., subtract CT from PT for every point. You can then plot the difference on the same time scale as the absolute powers (though on a different axis so that you can actually see the small values). This would show any systematic errors that might exist at certain power ranges or that might develop after a certain period of time. You could then also look at the average difference over the intervals, similar to what you have done already. As a partial aside, the average of a difference is not same thing as the difference between two averages. So the average of the point-by-point difference may not come out to be the same as the difference of the averages you have already calculated.

More than you wanted to know...:eek:🙂

Porkyboy said:

Hi

Looks Ok to me! :o

PBUK

Did you just change the values on the x axis? 😕

Hi Swampy

swampy1970 said:

Did you just change the values on the x axis? 😕

Of course, thanks for pointing out my error, twas getting late. Just having fun, I'd have confessed in a day or so! 😉

Cheers.

PBUK

Hi Steve

Steve_B said:

Just a small point about comparisons. I agree that you can tell just by looking at the plot that they are extremely close to each other and quite frankly, I think that you could just leave it at that; it's close enough for non-professional stuff.

Since you seemed to want to crank out some analysis though, I'd say it's helpful to compare the averages over certain intervals as you have done but overall it's better to compare the difference between the two on a point-by-point basis, i.e., subtract CT from PT for every point. You can then plot the difference on the same time scale as the absolute powers (though on a different axis so that you can actually see the small values). This would show any systematic errors that might exist at certain power ranges or that might develop after a certain period of time. You could then also look at the average difference over the intervals, similar to what you have done already. As a partial aside, the average of a difference is not same thing as the difference between two averages. So the average of the point-by-point difference may not come out to be the same as the difference of the averages you have already calculated.

More than you wanted to know...:eek:🙂

Thanks for the input and yes, lots more than I wanted to know! However, the analysis you suggest might be interesting as a one off. If you fancied it I'd be happy to send you the files and you could share the outcome. Bit too mathematical for me :eek:

Best rergards.

PBUK

Porkyboy said:

Hi Swampy

Of course, thanks for pointing out my error, twas getting late. Just having fun, I'd have confessed in a day or so! 😉

Cheers.

PBUK

Now, I know I've been little tired and a bit more confused than normal of late... but I was starting to think that I'd been seeing things after you posted your first reply. LOL

Steve_B said:

Since you seemed to want to crank out some analysis though, I'd say it's helpful to compare the averages over certain intervals as you have done but overall it's better to compare the difference between the two on a point-by-point basis, i.e., subtract CT from PT for every point.

That's actually quite hard to do, given, for a start, the different recording frequencies of each device, lining up the exact starting point in the data and dealing with any data drops or ways each device may treat data differently. Then there is what to make of PT's aliasing in such an analysis....

Sometimes just "eyeballing" it is sufficient.

Alex Simmons said:

That's actually quite hard to do, given, for a start, the different recording frequencies of each device, lining up the exact starting point in the data and dealing with any data drops or ways each device may treat data differently. Then there is what to make of PT's aliasing in such an analysis....

But the Chung method is easy? 😕🙂

I've done similar comparisons on TT files from two different riders with different recording frequecies doing the same course. You're right, it's not trivial.

Steve_B said:

But the Chung method is easy? 😕🙂

If it were hard, I wouldn't be able to do it.

Quoted post said:

I've done similar comparisons on TT files from two different riders with different recording frequecies doing the same course. You're right, it's not trivial.

If one is a PT and the other is any other brand it's not just not trivial, it's a pain in the butt.

RChung said:

If it were hard, I wouldn't be able to do it.

Just joking around. Thanks to your work and that of Adam and Alex (and anybody else), even I could handle it, if I actually get around to trying it. 😄

RChung said:

If one is a PT and the other is any other brand it's not just not trivial, it's a pain in the butt.

Fortunately both were PTs. Since it was a TT, I did the comparison by distance rather than by time so that the "features" would line up. I couldn't actually do a subtraction because the two riders had different average powers for the TT and used different sampling rates. So I compared the ratio of instantaneous power to average power between riders and also compared their velocities. Excel lets you plot using multiple x-data sources so the fact that one rider had 2x the number of samples was not a hinderance, at least to plot anyway. Employing a moving average helped make the plot more readable.

Steve_B said:

But the Chung method is easy? 😕🙂

I've done similar comparisons on TT files from two different riders with different recording frequecies doing the same course. You're right, it's not trivial.

This isn't the "Chung method" as it applies to virtual elevation/aero/rolling resistance analysis though. That's relatively easy, now that's the spreadsheets are out there 🙂
There is some stuff I'd like excel to automate for that analysis (like plotting laps on top of each other) but haven't applied my head to at the moment. Working on other stuff at the mo.

But yeah my original assertion was that a point by point comparison of files from different power meters is a PITA.

I'd like to know more about how WKO+ interpolates the data when you export files from a PT (or other meter) into an SRM .txt output (it creates a file with an apparent recording frequency of 1 second). Some while back when Dr Coggan showed me how to do an analysis of Maximal Accumulated Oxygen Deficit from pursuit files, that's how I pulled the data and it worked fine.

Alex Simmons said:

This isn't the "Chung method" as it applies to virtual elevation/aero/rolling resistance analysis though.

No, I understand. Totally different realm. I was just thinking of a spreadsheet that is a bit complex that many of knew about, that's all.