Hi, today I’ve noticed that NP of my ride on Intervals is different from NP on Connect, Trainerroad and Sauce for Strava.
Intervals 281 NP, 203 W avg
Connect 267 NP, 204 W avg
Trainerroad 268 NP, 203 W avg
Sauce for Strava 266 NP, 203 W avg
As far as I know the computation should be the same since it’s the same metric. Any ideas?
It’s not the same on Intervals. 30s rolling average taken which means the last 30s of your recovery intervals are included in NP of work intervals.
That’s a different thing. I am talking about NP for the whole ride. This one was 52 m 32 s. I’ve already checked the other threads about NP but none of them talks about this.
what’s the Moving time on all the sites? technically shouldn’t affect NP calcs tho. do you have quite a few stops / not moving etc?
It’s from a race so there’s just a few sec stop at the start.
This is related to the way Intervals.icu calculates the 30s moving average. In the beginning there isn’t 30s of data so Intervals.icu calculates a shorter average for each point until it gets to 30s. An alternative would be to start the calculation with 29s of zero before the actual data.
Usually this doesn’t make any difference. However you started that race really hard so this has over estimated the TSS for that part. The alternative would have underestimated TSS.
I am not quite sure what to do about this.
Oh that’s an interesting issue.
I think the mistake you do with the calculation is calculating the moving average before 30 seconds of activity. Instead you should start calculating the 30 sec moving average at the 30 seconds mark, so you have the moving average for 1s-30s, 2s-31s and so on, last interval ending at the last activity timestamp. Andrew Coggan describes the computation here.
Using this calculation I got 266, which agrees with the other sites.
I also tried removing all moving average zeros (0 power for 30 seconds) from the calculation and that gave me 268.
PS: Thanks for the option to extract the streams csv. Saved me parsing the file.
Tx for that info. I have fixed this. Intervals.icu now ignores the first 29 seconds. Will deploy in a few hours. You can re-analyse existing rides with very hard starts to get the new calc.
I think the best way should be applying a statistical distribution where based on the data this distribution takes away those upper and lower values considered “outliers” which are just exceptions that stand outside individual samples of population or in this case it may be whole population (data)
There’re many statistical distributions but the just applying a simple Normal Distribution, it effectively works out. Therefore, the “Normalized” power term comes from
For me Strava it’s the worst app to consider training or performance data because it take the elapse time as a norm which created a biased data no just for power but for speed as well.