mschwett 

did some searching on this, and there is understandably a LOT of discussion about how "accurate" strava's FTP numbers are.

a lot of the discussion is based on the fact that strava will estimate power without a power meter. e.g. this thread Strava FTP estimate

setting that aside, assuming a good power meter used correctly over a long period of time, does anyone know how strava actually makes this estimate? lots of people say it's the usual fraction of your 20 minute best effort, or 60 minute best effort, but that doesn't seem to be what it's doing in my case. it thinks my FTP is HIGHER than the strongest sustained hour i've ridden lately, and also stronger than the strongest 20 minutes i've ridden lately. is it smart enough to know that in an all-out, uninterrupted, well rested condition i'd do a little better? or is it also including performance on other durations, maybe longer ones? my power for short periods is very low, but pretty good out to 3 or more hours. does their algorithm care?

BTinNYC

Testing on the trainer about every 6 weeks I find Strava's estimates are very close to my test results. I see FTP as a training indicator not an on-road performance metric. So the number isn't very important to me, just the trend; increasing, flat or decreasing. In summary, be happy about 237 and shoot for 250!
All good, BT

Jughed

If your asking about Strava estimating without you having a power meter, in my experience - the strava numbers are way off,

My typical 17-18 mph ride is 180w+/- on my road bike with a power meter. My cross bike, same speed with fat tires and no power meter - strava estimates 135w for the effort. Not even in the same ballpark.

Even with my power data, for some reason Strava isn't 100% in sync with my Garmin. My last ride Garmin showed 200w avg, 215 normalized power (eliminating stops and times of 0 power) - Strava showed 168W. Sometimes max power is dead on, sometimes not.

Kai Winters

They are all marginally accurate but there so many variables that it should only be an estimate.

mschwett 

no, as i noted i’m talking about only with a real power meter! the power readings are accurate, but is the estimation of FTP …

RChung

That's interesting. I've never dived into the Strava FTP estimate so I don't know -- but it looks like you're using an 11-month data window? What happens if you reduce the width of the window? Does the FTP estimate change?

mschwett 

the window is just for the graph - whatever extent it bases the estimate on is unchanged by the length of time you view. EDIT : NOPE, wrong about that.

i set the window to that specific time frame because i made some methodology changes that slightly affected strava’s power curve display - and maybe the actual data. long story but the app that i use (cadence) stored milliseconds in the GPX file, which amazingly caused strava’s power over time curves to be way off, because the time values were scaled by 1.2-1.5x! the app maker updated the app to output whole seconds only and i reuploaded going back in time… then got tired of it. so i keep the curve display limited to the last 11 months. i don’t think the power estimate uses data older than that - but it could. i’d wonder if that could account for the higher value but it’s actually gone up over the time period, not down.

RChung

Is the FTP estimate being made by Strava or Veloviewer? I don't really use Strava but I thought the "power curve" was a subscription-only service?

mschwett 

actually, i have to take it back! i just fiddled with it and the estimated FTP does appear to be for the entire time window displayed - or at least, it's affected by the time window displayed.

the curve is from strava - i think it's part of the subscription plans.

here's the last 6 weeks vs the range i used previously. the relationship between the estimate the curves seems about the same - around 100% of 15 minute power, and 106% of 60 minute power.

Jughed

Do you have a garmin or wahoo type device? What are they telling you?

The Garmin connect app estimates my FTP and other metrics from the power curve/rolling data.

I know I could do a 20 min test and raise my FTP calculation - but I also know there is no way in hell I could ride a solid hour at my current FTP setting, let alone a higher one.

Our power curves are similar +/- and my calculated FTP is similar to the results you show.

Is the calculated data really accurate? I don't know.

RChung

Fascinating. So when you cut the window the estimated FTP changes but it still doesn't match the common (but wrong!) 100% of 60 minute or 105% of 20 minute. Do you ride with a HR sensor? Could Strava be using some other input than just power? I think Garmin says they use both power and HR (which seems odd to me).

mschwett 

yep, changing the timeframe of the graph changes the estimate, and definitely isn't 100% of 60 minutes or 105% of 20! i found as high as 253 and as lows as 223 depending on the window, which is in part explained by a different bike and PM for 2021, and the millisecond glitch for early 2022. but since fall 2022 the data is clean and always the exact same equipment.

i do ride with a HR sensor, always. my HR is very low, so maybe it's assuming i'm never actually trying hard!?!? or maybe it's also factoring in longer than one hour efforts.

RChung

Have you tried calculating your NP (or xP) over different durations and comparing that with Strava's estimate? Alternatively, have you examined your CP (and W') and compared it with Strava's estimated FTP?

mschwett 

looking at some stronger efforts with sauce, i see quite a few fairly uninterrupted 20 minute efforts with NP in the 235-250w range. it's hard to find a full uninterrupted hour, but as an example here's one with an unweighted average power (including perhaps 5 minutes of descending, intersecting, manuevering, etc) of 224w and an NP for that time of 236w. obviously my 1 hr FTP shouldn't be LOWER than an amount i actually output over an hour, so that sets a floor in the 220s. an average HR (as below) of 122 is VERY high for me, approaching maximum effort. so maybe strava's current 237w is a pretty good number.

mschwett 

i do not actually, the app i use has great dashboards, routing, seamless strava sync and ride history, but it does not draw any conclusions about FTP from results over time.

Jughed

Max HR of 136!! You must have an elephant sized ticker.

RChung

It would be sort of surprising if Strava were using NP as part of their FTP estimation algorithm, but it does seem closer than a fixed percentage of 20 minute or 60 minute power. You know how the estimated FTP varied with window size? If you narrowed the window way down (and the FTP changes) does the max NP still come close to the new estimated FTP?

mschwett 

that's not a good thing lol. and 136 is actually 6 over my target max! most rides i hang out between 95 and 125 for 95% of the time.

mschwett 

the narrowest it'll let you make the power curve window is about a week, and the FTP jumps around by about 20 watts as i move it from week to week. can't see any correlation to the NP of the strongest hour in the strongest ride in that week. they're always within 20 watts though, e.g. FTP 227 NP 237, FTP 229 NP 234. here's 6 days that had three strong rides, one with a decent hour, again, some descents and maneuvering, but just a few minutes, with a particularly close correlation : 235 FTP and 233 NP for an hour.

mschwett 

i think my preliminary conclusion is that they're using more data than just the 20 minute or 60 minute average power or normalized power for any one ride in the time period, but that the number is not far off from the true straight-average power i could comfortably sustain for an hour or more in ideal circumstances. so, a useful number to set zones by.

RChung

That's a good outcome, but now you have me wondering how Strava does it.