New WKO

gilliatt

 Is this a new WKO 3 thing I've been noticing some trainer intervals where my average power is higher than my normalized power.  I was a 2.2 user until real recently.  

 

Here is an example.

 

nterval 3:

Duration:   5:08

Work:       82 kJ

TSS:       9.6 (intensity factor 1.062)

Norm Power: 265

VI:         0.99

Pw:HR:       n/a

Pa:HR:       n/a

Distance:   2.277 mi

Min Max Avg

Power:       84 347 269 watts

Cadence:     72 100 89 rpm

Speed:       23.2 29.8 26.5 mph

Pace         2:01 2:35 2:16 min/mi

Crank Torque: 82 346 255 lb-in

 

mancona

I've noticed the same on 5-10 minute intervals where my vi is .99. Supposedly, they update the formuals in 3.0 to be "more accurate".

IIRC, Andy Coggan has said that for around 5 minutes or less NP is not accurate and you are better off using Avg anyway.

In your example above I would just say you held 267 watts for the 5 minutes and not worry about the small difference.

William Jenks

RaceDay gives result like that as a result of doing a 30 second exponential averaging rather than the 7 second moving average that WKO 2.2 does. For a short interval (like 5 minutes), the averaging on the front end used in the NP calculation (xpower in RaceDay) includes pre-interval power, which is typically lower, but doesn't give you any "credit" at the end, so it lowers the NP a bit. As the intervals get longer, it doesn't do that any more because the 30 seconds becomes insignificant.

I wonder (do not know) if WKO 3 just changed their averaging scheme vis a vis 2.2.

You have to do SOME averaging to get NP (or its xpower equivalent) because if you don't ,then you end up squaring the anomolously high points and throw off your distribution too much. (think about getting the sum of the squares of the completely unaveraged data vs that of the smoothed data...obviously the square of the big numbers grows faster than the square of the smaller numbers, so it makes the noise problem worse)

William

Coach Patrick

@William, I understand what you are saying but have no idea what you just said. I think it's negligible in 5' increments...

William Jenks

Bottom line: I know what Race Day does, but I don't know what the New WKO does. WKO did state that they changed how they handled the data averaging for TSS/NP, and I believe the effect of what they did is to make the rolling average longer. However, I think it's pretty clear what's going on.

The rolling average (or exponentially weighted one, in the case of RaceDay or Golden Cheetah) means that if you do a "step function", i.e., suddenly increase your effort and push the button to start counting it right then... then the "rolling average" takes a little while to catch up with you, as it throws away the old data at the low power.

In any fitting procedure, how much (and how) to smooth the data is a big question. Here, they've taken an approach that makes the data take as long as half a minute to "catch up" to your current power output. So if your interval is very long compared to that half a minute, it doesn't matter and the benefits of the data smoothing outweigh the bad part, but if your interval is only a few minutes long, this process distorts what you did. I wish I had a piece of paper to sketch this out.

Maybe one way to think about it is to compare the "instant" meaurement of power to the delayed response of HR (which is much smoother than power!). Your power goes up, and your HR catches up over the next minute or so. With a moving average (or exponentially weighted average), the same phenomenon happens, comparing the actual power and the averaged power.

Oy. I wish I were better at explaining things with just typed words.

steveperx

 I think we need some charts and graphs.  Maybe a whiteboard movie? 

mgraffeo

Maybe think about it this way:

-Another way to say "averaging" is to say "smoothing". Take a power file with lots of spikes, and you smooth that out in order to do most of these calculations (so that one stomp on the pedals of 1000 watts doesn't really throw everything off). There are different ways to smooth data,and those will generate different numbers called "average".

-WKO apparently changed from one to another. Therefore, under certain circumstances, the average will come out differently.

-which one is right? Impossible to say. There isn't a real way of measuring "training dose", so we do the best we can with these guesses.

-does it matter? Only for very short intervals (under 5 minutes), and even then, not so much.

So, takeaway? Don't worry too much about it...

:-)

cwhyte

I'm in total agreement with Mike G. Here's a good post to read:

http://forum.slowtwitch.com/gforum.cgi?post=1813136;#1813136

To sum it up, it's good enough. We don't execute at a very specific power level. It's always a range and where we should likely sit within that range depends on several factors. We just need something to give us enough guidance that allows our RPE to do the rest, imho.

Thanks, Chris