Ergomo Test Results

Bruce Diesel said:

I am in the process of trying out an Ergomo, I am a PowerTap user and a Computrainer user. So I set up my bike with both devices, put it on the computrainer and created an erg file that stepped 50 watts every 30 seconds, from 150 watts to 350 watts. Attached is the graph of the results - pretty good.

My mistake was to not run the test for long enough to be able to compare NP for the two files - will redo!

Do you have the Normalized Power (NP) for the two?

Which of these does everyone think is he closest to reality?

TF

Terry Ferguson said:

Do you have the Normalized Power (NP) for the two?

Which of these does everyone think is he closest to reality?

TF

No, CyclingPeaks needs a minimum of 5 minutes (I think) duration before it will calculate NP - hence no NP number.

I think the important thing as that both devices are very consistent and within a few watts of the reference device (a computrainer in this instance).

Bruce Diesel said:

I am in the process of trying out an Ergomo, I am a PowerTap user and a Computrainer user. So I set up my bike with both devices, put it on the computrainer and created an erg file that stepped 50 watts every 30 seconds, from 150 watts to 350 watts. Attached is the graph of the results - pretty good.

My mistake was to not run the test for long enough to be able to compare NP for the two files - will redo!

Bruce,

Just one suggestion:

Please rethink the coloring of the lines...

That would really help the color blind amongst us... 😄

I am having a real hard time deciphering which line is which...

Jim

otb4evr said:

Bruce,

Just one suggestion:

Please rethink the coloring of the lines...

That would really help the color blind amongst us... 😄

I am having a real hard time deciphering which line is which...

Jim

Sorry about that, just took the defaults that excel threw at me.

Okay, the smoother one is the ergomo.

Bruce Diesel said:

My mistake was to not run the test for long enough to be able to compare NP for the two files - will redo!

You can manually compute NP.

Bruce Diesel said:

Sorry about that, just took the defaults that excel threw at me.

Okay, the smoother one is the ergomo.

Not a problem...

Thanks,

Jim

Bruce Diesel said:

Okay, the smoother one is the ergomo.

Why is the Ergomo-Line smoother? Less acurate or simply a rolling average?

carmol said:

Why is the Ergomo-Line smoother? Less acurate or simply a rolling average?

Suspect the Ergomo uses a 5 second rolling average - not sure though.

RapDaddyo said:

You can manually compute NP.

Okay, got the pocket protector out so here are the results

Powertap: AP = 198, NP = 232
Ergomo: AP = 189, NP = 221

Would anybody care to state whether these differences are significant?

Bruce Diesel said:

Okay, got the pocket protector out so here are the results

Powertap: AP = 198, NP = 232
Ergomo: AP = 189, NP = 221

Would anybody care to state whether these differences are significant?

Considering that:

1) the Ergomo should read higher, not lower, than the PowerTap (since it is "upstream" of chain), and

2) training-induced changes in power from the nadir of out-of-season fitness to the peak of in-season fitness might be only 25 W in some individuals

then I'd say that they are significant (in the non-statistical sense of the word).

If you want to test the significance of the difference statistically, I'd suggest time-aligning the data as best you can, then plot the power data against each other and see if slope and intercept of the regression line deviates significantly (P<0.05) from 1 and 0, respectively.

acoggan said:

If you want to test the significance of the difference statistically, I'd suggest time-aligning the data as best you can, then plot the power data against each other and see if slope and intercept of the regression line deviates significantly (P<0.05) from 1 and 0, respectively.

The data is time aligned as best possible as it stands. Don't know if the pocket protector can cope with the stats though - will check it out :eek:

acoggan said:

If you want to test the significance of the difference statistically, I'd suggest time-aligning the data as best you can, then plot the power data against each other and see if slope and intercept of the regression line deviates significantly (P<0.05) from 1 and 0, respectively.

That method isn't robust against even small errors in time-alignment.

RChung said:

That method isn't robust against even small errors in time-alignment.

If you have a better approach (other than the eyeball test, which IMHO doesn't hold water), I'm all ears.

acoggan said:

If you have a better approach (other than the eyeball test, which IMHO doesn't hold water), I'm all ears.

It's pretty clear from the graph that the Ergomo is time averaged (pretty sure the Ergomo uses a 5 second rolling average) where the PT has more variability. How about smoothing both data sets by say a 10 second to 15 second windowed average. That would tend to minimize small time misalignments but should still demonstrate correlation or lack thereof. A 10 second averaging window still gives 3 samples per each 30 second power step so your results should be valid from a sampling theory standpoint.

Eyeballing the data it looks like the two meters track pretty well at lower powers and diverge from the midpoint of the ramp test through the higher powers. If the difference between the two is a constant percentage you could just bump up the calibration coefficient of the Ergomo but if they actually diverge then you're hosed at one end of the scale or the other. Course it's hard to say which meter is in error or which might be suffering a tracking error unless you can also show the third curve from the computrainer.

-Dave

acoggan said:

If you have a better approach (other than the eyeball test, which IMHO doesn't hold water), I'm all ears.

Your HO is your HO, but here's a test: take any one of your ride files. Copy it, but in the copy introduce a one-second hiccup halfway through. Do your comparison of the two files using your method, and with the eyball method. Does your method reject the null that the slope is 1 and the intercept is zero? If so, then your method isn't very robust against small mis-alignment errors. Unless the file you chose was quite unusual, I'm guessing it will reject. (I'm also guessing that the eyeball test will show the hiccup, but that's secondary).

RChung said:

Copy it, but in the copy introduce a one-second hiccup halfway through.

Alternatively, append a single zero record to the end of the original file and add a single zero record to the beginning of the copy, as if you were starting two measuring devices one after the other, and ending them the same way. The two files will have the same lengths, AP, NP, TSS, and variability. QA will show the same things. The training implication will be the same. Can you reject the null that the intercept is zero and the slope is one? If so, then I'd say that your method isn't robust to small errors in time alignment.

daveryanwyoming said:

It's pretty clear from the graph that the Ergomo is time averaged (pretty sure the Ergomo uses a 5 second rolling average) where the PT has more variability. How about smoothing both data sets by say a 10 second to 15 second windowed average. That would tend to minimize small time misalignments but should still demonstrate correlation or lack thereof. A 10 second averaging window still gives 3 samples per each 30 second power step so your results should be valid from a sampling theory standpoint.

Eyeballing the data it looks like the two meters track pretty well at lower powers and diverge from the midpoint of the ramp test through the higher powers. If the difference between the two is a constant percentage you could just bump up the calibration coefficient of the Ergomo but if they actually diverge then you're hosed at one end of the scale or the other. Course it's hard to say which meter is in error or which might be suffering a tracking error unless you can also show the third curve from the computrainer.

-Dave

I'll prepare a graph with the computrainer data included. The first time I did this, I had an Ergomo that was significantly out ~50 watts. Fiddling the k-factor created some really wierd results - non-linear behaviour.

I am working with the Ergomo Agent here in South Africa - he has asked me to try out the Ergomo so we can learn a bit about it. With the first meter, the guys in Germany felt that the installation was not correct - and in taking the BB back out the mechanic slipped and sheared off the wires - the unit is sealed so we had to write that one up to learning about the device - and send it back to Germany. The graph I have posted above is with the second unit, installed as accurately as possible and using the k-factor that come on the calibration certificate.

From my initial test runs - the computrainer and powertap were spot on - so I have a lot of confidence in the PowerTap calibration. The computrainer curve is also a lot smoother than the powertap, so maybe they use a rolling average as well.

Will post the computrainer later today.

Okay,

Have attached two graphs:

1. PT vs CT
   
2. Erg vs CT
   

As you can see, PT tracks the Computrainer almost perfectly. Computrainer also appears to be running a rolling average.

Have tried to make sure that they are easy to see 😉

RChung said:

Your HO is your HO, but here's a test: take any one of your ride files. Copy it, but in the copy introduce a one-second hiccup halfway through. Do your comparison of the two files using your method, and with the eyball method. Does your method reject the null that the slope is 1 and the intercept is zero? If so, then your method isn't very robust against small mis-alignment errors. Unless the file you chose was quite unusual, I'm guessing it will reject. (I'm also guessing that the eyeball test will show the hiccup, but that's secondary).

You haven't answered my question.