Strava is a wonderful social media site for endurance athletes to share their rides and runs, praise each others’ efforts, and try to poach the best times on specific routes around town.
I’ve been a fan of Strava for a long time, as are hundreds of thousands of other cyclists, triathletes and runners. It’s mind-boggling to see the power output that pro cyclists and triathletes post there. Even before I had a power meter, it was neat to see what Strava estimated my output to be during a ride.
But I wondered just how accurate these power estimation algorithms are. After all, GPS data is less accurate and doesn’t account for the wind.
So I tested Strava by comparing my power meter’s wattage data to Strava’s estimate of my power that was based only on GPS data. You can read about my first test results here: “How Accurate Is Strava? Is Strava as Accurate as a Power Meter?”
All in all, Strava performed fairly well. But my test was on a flat course with multiple laps through brisk-yet-constant winds. For all intents and purposes, the site was a race car track. It was about as controlled an environment as you can get outside of a wind tunnel. It was much shorter than a typical Olympic triathlon bike course, let alone a 70.3 or Iron distance race. So while the Strava algorithm passed, it hadn’t quite proven its versatility. I decided to give it another go, this time over a hillier route with multiple changes in wind and road surface. Instead of repetitive laps, it was roughly 35 miles out and back. This second test is more “real world” riding. Here are the charts.
Then the data from Strava:
The test method was the same as before: I uploaded the ride data simultaneously to Golden Cheetah and Strava. Then, using the Golden Cheetah ride editor I removed all power data and fed it into Strava as a different ride to allow it to estimate the power output. I compared the Strava outputs for the overall ride and ride segments. To give Stava’s estimator the best possible chance I weighed myself and my bike, fully equipped and with full water bottles, before starting the ride. I changed the appropriate weight data before uploading the rides.
Given measured power data, Strava said that my average power for the whole ride was 164 watts and produced a total of 1,233 kilojoules of energy. Without direct measurements, Strava estimated that my average power was 177 watts and total energy production was 1,328 kilojoules. That’s an overall power error of 7.9%. That’s not bad—but it is more than double the error of the previous test. This indicates the power estimation on Strava gets shaky as the course accumulates variation. You can see this happening visually when you zoom in on the performance graph for a shorter segment.
Here’s the data from my power meter:
And then here’s the data from Strava:
Visually speaking, there’s a lot more “chop” to the estimated data. This makes sense. Because the estimation process uses speed and weight compared to GPS distance instead of the power meter’s fixed sampling rate, the Strava estimation is going to miss moments of power variation.
Breaking it down further, there were 69 unique user-designated segments on the route I took. On a per-segment basis, the error in power estimation was 15.5%, which was double the previous test. Remarkably, the average length of the segments in this test was very close to those in the previous one, so the length of a Strava segment does not seem to be a factor in creating error.
Drilling further into error on individual segments, the greatest error was 34.4% and the lowest was 1.9%. Thrown out of this data set are two outliers that also reveal something informative. In those cases, I rode the segment at an average output of 20 watts or less. When that happened, Strava’s error was 120% and 400%. I discount those because they are exceptionally low wattage outputs and not worth considering. I suspect that Strava’s power algorithm is intended to work within a range of “useful” power outputs. On a segment where my average power was under 100 watts, the estimator error fell to 21%, so that range is pretty versatile. Strava seems to have trouble with estimating very low efforts.
Strava overestimated power on 53 of 69 (76%) of segments, which suggests it will routinely pump up your ego (ironic, Strava, and diabolical!). I could find no apparent correlation between a specific factor and error, though. It’s possible that winds played a much more significant role on a course that didn’t loop, but I can’t eliminate the possibility of pavement conditions, either.
Climbers will be happy to learn that the Strava power estimator was very accurate on the climbs (a few category 4s and a category 3, per Strava’s mapping). Other segments with larger error had no distinct feature to suggest the cause.
In conclusion, it seems that Strava’s power estimation is in fact probably more helpful to triathletes than dedicated cyclists. If you have a favorite training ride that’s fairly repetitive—whether laps around a short, flat course or hill repeats—you’ll get accurate estimations of your performance.
Why might Strava’s estimation be better on hills or repeated circuits? There are two reasonable answers. First, repeating hills or riding the same loops several times gives Strava the opportunity to average your performances and come up with a more accurate, apples-to-apples estimation. Second, these short courses help neutralize road and wind conditions, which removes some uncertainty from the data.
But if you’re going for a long, meandering weekend ride, expect your data to be less helpful the farther afield you go. Longer rides give you more opportunities for your performance to be influenced by the wind and other conditions, which Strava’s power estimates will miss.
Until you can actually measure and see your power output in real-time on the bike, athletes should consider Strava’s post-workout analysis to be useful mostly for post-workout bragging (a valid and entertaining use, to be sure!). Once Strava’s power estimates have helped you get familiar with your relative power output in different situations, you should consider investing in a power meter for more accurate, real-time data on your performance on every ride, in all conditions.
On a final note, those of you watching your calories may be interested to know that the estimation errors will spill over into this data as well.
Calorie data from my power meter:
And calorie data estimated by Strava:
If you’re interested in getting faster, you’ll be fascinated by FASTER: Demystifying the Science of Triathlon Speed. In Faster, astronautical engineer and triathlon journalist Jim Gourley explores the science of triathlon to see what truly makes you faster—and busts the myths and doublespeak that waste your money and slow down your racing. With this knowledge on your side, you can make simple changes that add up to free speed and faster racing.