How can NP be lower than avg power?

Title, i did Baird +2 outside today and I got a 10 W lower NP than avg power. Is this a bad sign?

Overall, or on the intervals?

Overall, you can’t really get lower NP than average power. So if you did, it sounds like some kind of bug.

On the intervals (photo below)

well, is possible. In case if you use e.g. Auto Pause. Then the zero values not affect the avg power, but NP

ok, here the low values before an after the interval affects NP. On the other hand the NP of breaks increases

I didn’t stop at all during the intervals or recovery.

yes, but second part i wrote is valid. Here the NP algorithm

Normalized Power

Normalized Power (NP) is a metric to quantify training intensity with power data and is introduced by Andrew Coggan. The concept of NP is discussed in chapter 7 of the book. It is especially useful in conjuction with the other algorithms below.

Step 1: Calculate the rolling average with a window of 30 seconds: Start at 30 seconds, calculate the average power of the previous 30 seconds and to the for every second after that.

Step 2: Calculate the 4th power of the values from the previous step.

Step 3: Calculate the average of the values from the previous step.

Step 4: Take the fourth root of the average from the previous step. This is your normalized power.

So in the first and last 30 second, the NP is affected by the power in the break.


Yes. The first step in computing Normalized Power is smoothing the power data with a 30-second average. TrainerRoad uses this smoothed power from the whole activity for computing normalized power for segments (intervals, selected regions, etc.). So the first and last 30 seconds of any segment are averaged with some power data not in that segment. For short segments, that’s a big fraction of the segment. If you’re doing intervals, so what’s adjacent to a lap is rest, that will really pull down the average. Since they use the unsmoothed power data for computing average power, it’s not affected. As a result, you’ll easily get higher AP than NP.

All it means is that you can’t use normalized power for short intervals.

1 Like

Yup, NP should pretty much be ignored (or considered as purely “interesting”) for short intervals. They covered it on the cast recently, and said something like 5 minutes is about the lowest time you can go and expect anything approaching meaningful. But I think the common recommendation is closer to 20 minutes being the real minimum.

1 Like

The rule is, whichever is higher, that’s the real power you should pay attention to.


Normalised power should be ignored for anything less than about 15 to 20 minutes.

It’s trying to equate the physiological cost of a variable effort to a steady state effort.


1 Like

Are you excluding the zero’s in your power data recording?

Which means this 11 minutes of Donner:

Is roughly equivalent to this 11 minutes of Mills -5:

Because 225W is about 231W.

The point being, normalized power allows you to compare:

  • variable efforts (hills or on/off intervals)
  • steady efforts (flat time trial or on interval)

You can’t use average power to do that - 226W vs 193W for those 11 minutes of Donner vs Mills-5


That is, in essence, exactly what NP is trying to do. However, at 11 minutes your example is a bit on the short side.


If NP is the same, then TSS is the same (assuming same FTP).

In general, it is just fine to compare efforts even at short durations. Is that useful info? Separate question.

NP doesn’t incorporate muscular endurance, energy system contribution, etc. Therefore someone that can do an 11 minute set of vo2max intervals can probably do the “NP equivalent” threshold segment at say 96% steady state for 11 minutes (my example above).

But say you do 250W normalized power for an hour hilly race, you may find it difficult or impossible to go out next week and do 250W steady-state for an hour. Which brings to mind “not all TSS are created equal” which is equivalent to saying “not all NP are created equal.”

Just look at NP and AP at the end. During the activity both of those are meaningless and just follow 3-10s power.

Surely it has to be useful to make it worth looking at in the first place, otherwise what are you getting out of it?

The WKO4 FAQ page is clear that the physiologic toll is not equal to that of a steady state effort until after 20 minutes: “The measurement doesn’t become an accurate representation of the physiological toll until at least 20 minutes.”

Shame the very first line of the article has an error…


Sometimes understanding when a model fails is just as important as understanding when it works well.

Thought this 3-part blog series did a good job explaining the ins and outs of NP vs Average Power:

1 Like

Good articles.

When referring to “what are you getting out of it?” I was specifically talking about useful data, rather than useful knowledge, and as such I don’t think there is useful data that you can get out of looking at NP for durations under 20 minutes, as stated in Part II.


Just finished my first 20 minute FTP test and learned a couple things: normalized power can be lower than average power even for a 20 minute interval (2 watts lower in my case), and TR uses .95 of normalized power for calculating my FTP, not average power. Since the result is that my new FTP ends in a 9 instead of a nice round number you can imagine I’m devastated :slight_smile:

Oh yeah I also learned that I like the 20 minute test better than the ramp test :slight_smile:

Edit: I spoke with TR support and they said the way it calculated my FTP after the 20 minute test “should not be happening.”

Is anyone aware of any studies that demonstrate the validity of NP as a representation of physiological toll? It’s common in conversations for us to say NP allows for comparison of effort across more variable workouts, but what is that based on, really?