Latest revision as of 15:06, 30 December 2017

Two Running Sensors that estimate running power, the Stryd on the left and SHFT on the right.

I've tested many of the running power meters that are currently on the market. While some of them are remarkably useless, some of them can provide value as long as you understand the limitations.

1 Limitations of Running Power Meters

Here are the primary limitations of running power meters, and different sections of this article will go into more detail on some of them.

• They are not "power meters" in the sense of actually measuring power. Running power meters estimate running power by modeling a runner using inputs such as pace and incline (see below for details.) This means that they are only a rough estimate, unlike cycling power meters that actually measure power directly (again, see below for details.)
• A key input to any running power estimate is your speed. That means they are dependent on whatever mechanism they using for measuring your running pace. Any system that relies on GPS for pace will be deeply flawed under many circumstances.
• The second major input to a running power estimate is incline or decline, and modern pressure sensors can give a good estimate of slope. However, they don't tend to work well for slight angles, and most systems seem to include a fair bit of smoothing, so the response to elevation changes is slower than a Footpod can detect pace changes. This can produce some art artifacts when you run up a short steep slope for instance. Often, the power estimate will detect you slowed up, and drop your power estimate quite low, then when you're back on relatively level ground it will notice the incline change and combine it with your level running pace, producing a really high estimate of power.
• Because they are not measuring power, they provide no useful insight into Running Economy. This seems to be a common misunderstanding.

2 Are They Useful?

There's an old saying that " all models are wrong, but some models are useful." So, while running power meters can't measure your power output, they can still be useful. It's possible for a running power meter to estimate how an incline or decline changes exercise intensity for a given pace. This can allow for more even pacing on hilly courses, which gives a more even effort when training. More importantly, it can allow for more even pacing on a hilly race, which would be great for things like the Boston Marathon. This could be more effective than Heart Rate, which suffers from a significant lag between changes in intensity and a change in Heart Rate, as well as Heart Rate Drift. A running power meter can also be useful when doing uphill interval training, where they can provide some insight into the effort required compared with level ground.

3 Power Estimate or Heart Rate?

The use of heart rate for training has been established for many years, and heart rate based training has some useful advantages, as well as some significant shortcomings. I think the estimate of power output overcome some, but not all of the issues with heart rate based training. Personally, I don't see this type of power estimate completely replacing heart rate based training, only augmenting it.

• Heart rate responds to exercise intensity with a delay, while power estimate is much closer to real-time.
• During longer exercise, Heart Rate Drift occurs that generally causes a higher heart rate for a given intensity. The reasons for this drift are complex, and include dehydration, fatigue, carbohydrate depletion. Using a power estimate ignores this drift, though it's unclear to me when to use heart rate and when to use the power estimate. In some situations, it seems likely that the drifted heart rate is a better estimate of intensity than an unmodified power estimate.
• There is a widespread myth that Maximum Heart Rate can be calculated, leading to some erroneous assumptions of how a given heart rate relates to the percentage of exercise capacity. In practice, both Heart Rate and maximum estimated power require a practical test.
• A common use of heart rate data is to allow an athlete to train at their Lactate Threshold, often referred to as Tempo Runs. The belief is that this training intensity is especially beneficial, though the available science indicates the opposite. If Tempo Runs made sense, then a power estimate would be quite valuable for hitting that pace accurately. I'm sure that many runners will use power estimates this way, even though the science indicates it is ineffective.

4 Running Power Meters Tested

I've tested a number of running power meters, and they vary vastly in the usefulness.

• The best I've found so far is Stryd which gives some usable estimates of power. It has one of the most accurate estimates of running pace, which is the primary input to a power estimate. I've found it gives a pretty good equivalence between level ground and running uphill, but seems to underestimate a little on the downhill's. However, I suspect there is far more individual variability when running downhill, as different runners will have to break more to keep their pace lower where other runners will be able to speed up more easily. The lag between a change in effort and the Stryd change in our estimation is usually fairly short, measuring only a few seconds. Not surprisingly, it's more responsive to changes in pace than changes in incline, and it doesn't always pick up very well on extremely shallow angles that can still make quite a bit of difference to your running effort.
• Garmin has released their Garmin Running Power as a Connect IQ app that I'm testing now. My suspicion is that this approach will depend largely on the accuracy of the pace measurement. Therefore, I'm expecting pretty useless results from GPS, but more useful information when the Garmin watch is paired with a Footpod.
• The SHFT sensor is fairly useless in its power estimate. It doesn't seem to be able to estimate pace very well, and is particularly useless on the downhill.
• I'm waiting for RunScribe Plus to reach a functional state. So far, it's not stable enough for me to spend time testing it.
• The Garmin Connect IQ platform should allow for alternative models of power, given pace input from an accurate Footpod and the barometer of the watch.

5 How Do They Work?

So far, only RunScribe has been transparent about the model they use for running power, which is available as an online calculator at RunScribe.com. Your power output is going to be closely related to your Running Economy, and there's quite a bit known about The Science of Running Economy. The three most important inputs to the model are:

• Weight. A key input to any estimate of running power is your body weight, and generally running effort varies linearly with your weight. This is pretty easy to estimate, or running power systems ask you to enter your weight. I typically enter my approximate body weight and then don't modify it, but you could weigh yourself before each run in your running gear to compensate for additional closing and the like. Of course, it's hard to adjust for hydration changes; in other summer my body weight can vary by 5Lb/2.5Kg over the course of a long run. None of the running power systems evaluate the weight on your feet, which we know can dramatically influence Running Economy.
• Pace. For running, effort varies linearly with your pace across a broad range. Accurately measuring your pace is one of the critical problems in estimating power. Some systems will rely on GPS, but GPS Accuracy is so poor that the systems tend to be deeply flawed. A better approach is a Footpod, and systems like Stryd don't need calibration to produce high accuracy.
• Incline. As everyone knows, running uphill is hard up and running on the flat, so an accurate measure of incline is critical. So far, I'm not aware of any system that using inertial guidance to measure incline. All of the systems rely on a barometer to measure pressure changes. This means they tend to be pretty good when moving up a reasonable incline, but they don't always notice a shallow incline that's still enough to make a difference to the effort required. A bigger issue is understanding the effort required when running downhill, as this is not such a simple relationship between effort and angle. Initially, Stryd did an awful job of estimating the effort of running downhill, but subsequent firmware updates greatly improved this.

There are some other possible inputs, but they appear to be tricky to implement at best.

• Air resistance. RunScribe use height and weight to estimate your frontal surface area to better take into account the contribution of air resistance to the overall effort of running. Obviously, the wind makes a big difference to running effort, especially a strong wind. Unfortunately, it's not practical to measure wind speed on the runner, though the Stryd pattern suggests doing this using a barometer. (I've no idea how that would work, as you need to know direction as well as speed.) Garmin attempts to overcome this by using a cell phone connection and GPS to get the wind speed and direction from the local weather station. This seems like a bad idea to me, as the wind speed and direction at the weather station could be vastly different from where your running. For instance, I typically run on a sheltered Greenway that rarely gets any air movement, even when the weather report is indicating it's breezy.
• Cadence. There is some evidence that Cadence influences running economy, and therefore could be an input to an estimate of running power. However, the research doesn't seem to be well enough and defined that this is easily generalized in a way that would allow it to be modeled.
• Vertical Oscillation. It's intuitively obvious that the up and down movement of your body reduces your efficiency compared with minimal vertical movement. Unfortunately, the science doesn't backup this intuitively obvious observation, and is a relationship between vertical oscillation and running economy is unclear. This is likely to be due to a runner having some elastic properties making them a little more like a bouncing ball. In one study, reducing vertical oscillation reduced running economy, though findings are not consistent.
• Ground Contact Time. There is some evidence that faster runners have lower ground contact time, but the relationship between ground contact time and running economy is mixed at best.

6 Running Power and V̇O2max

A runner's pace is directly and linearly related to their oxygen consumption across a fairly wide range of running paces. And pace is a primary input into running power estimates, which has led some to conclude that the running power estimate is a good substitute for estimating absolute oxygen consumption. This doesn't work for a number of reasons…

6.1 Overall Power

Even if it were possible to accurately measure running power (which it's not), this would still not be a measure of oxygen consumption. Running power is a measure of output, where oxygen consumption is a measure of input. The difference between the two is efficiency, and most of the oxygen you consume running generates heat. You can estimate your power generation from your V̇O₂max:

• You can estimate your V̇O2max from a recent race time.
• Given a 3-hour marathon, that would be V̇O2max of ~54 ml/kg/min.
• You can convert oxygen consumption to watts. 1 MET = 3.5 O₂ mg/kg/min = 1.16222 W/Kg.
• So, 54*1.6222/3.5 = ~18 W/Kg
• For a 140Lb/63Kg runner (me), that's 1,340 Watts of total output.
• My actual power at V̇O₂max is vastly less than 1,340 Watts.

Most of that power output is not running power, but goes to generate heat and do other things superfluous to running like keeping my brain working.

6.2 Metabolic Efficiency

The conversion factors used above also only work if you are burning carbohydrate, as fat burning requires significantly more oxygen. It would be wonderful to gain insight into the mixture of fuels that an athlete is burning, but so far, the only way of doing that is to measure the ratio of oxygen consumed to carbon dioxide produced.

6.3 Running Economy

One of the primary benefits of training is likely to be improving running economy, so that you can run faster for the same oxygen consumption. Running economy varies significantly between different runners, which means that running power output would vary for a given oxygen consumption between runners. If a running power meter was actually measuring oxygen consumption, then this would be a way of measuring running economy, but that's an equally flawed idea.

6.4 Comparisons Between Running Power and V̇O2max

Below is a graph from the book "The Secret Of Running", which compares power output and V̇O_2. It's claiming that this shows "the stride data I just as good as the V̇O₂." However, this only shows that the Stryd estimate of power varies linearly with pace, which given pace is a primary input to power estimate should come as no surprise. It also shows that oxygen consumption is also linear with pace, something that is well known. However, the power estimate for a given oxygen consumption varies dramatically. You can see for 40 ml/Kg/min, the estimate varies from 2.5 w/Kg to 4.0 w/Kg, a nearly 40% variation. See hetgeheimvanhardlopen.nl for more details.

Data from "The Secret Of Running" book.

Stryd have released their own evaluation at Stryd Metric Validation. Their data shows "Stryd power is 96% correlated with metabolic energy expenditure", but the study was on 13 runners and the correlation is only a subset of 9 of them.

Data from an evaluation performed by Stryd.

7 Differences Between Running and Cycling Power Meters

Power meters have become an integral part of training for cyclists, and a number of running Running Sensors claim to have similar benefits for runners. Let's look at the differences between power for cycling and running.

• The amount of energy required to run a given distance on level ground is fairly constant, which means that for runners, pace is often an excellent estimate of effort. By comparison, the effort required for cycling varies with the square of speed and is impacted far more by wind speed and aerodynamics, so speed is fairly useless.
• For cyclists, it's easy to define power output, as it's the energy used to propel the bicycle forward. For runners, it's far less obvious what should be considered power. You could consider it purely to be the horizontal force applied to the ground, as that's the only power that goes towards forward movement, but that's only a tiny fraction of a runner's energy expenditure. Much of a runner's energy expenditure goes to vertical movement, but some of that movement is elastic balance, and some of it requires energy. That means that defining power output for a runner is tricky at best.
• It's quite easy to measure cycling power by measuring the force applied (torque) and the rotation it's applied through. This makes cycling power meters reasonably accurate and not outrageously expensive (compared with say, a metabolic cart that a lab would use to measure oxygen consumption). By comparison, running power is impractical to directly measure and requires a mathematical model with various assumptions that use indirect inputs. This makes running power meters an estimate of power output rather than a true meter. If cycling power meters worked like this, they'd be a wind speed meter that would guess how that influence power output.
• Because running is vastly less efficient than cycling, knowing power output is less important. If a cyclist knows their power output, they have a very good estimate of their exercise intensity. This is because most of a cyclist's effort is delivered as usable power. Even if it were practical to measure running power, it would be of relatively less value due to the influence of Running Economy. This means that a runner is far more interested in their energy expenditure than of their power output.

8 Running and Cycling Watts

Garmin claim that for a given oxygen consumption, the Watts of running output should be higher than watts of cycling^[1]. Thankfully, Garmin list the scientific research they use to support their claim, allowing for their conclusions to be verified.

• Claim: Metabolic efficiency of positive work for running at 2.75 m/s is 39%, and at 3.25 m/s is 41%^[2].
  • This study uses optical measurement of joint dynamics along with ground reaction forces to calculate the total mechanical power, and compared this with the oxygen consumed. The focus of this study is to estimate how much power is being generated at each joint, rather than overall efficiency. This study ignores any elastic properties of a runner, and assumes that none of the landing forces are used to subsequently generate positive forces. It's debatable whether this is a reasonable assumption given the limited focus of this study, there is plenty of evidence to the contrary^[3][4], with one study claiming "In walking or running the negative work performed in the eccentric phase was completely stored as elastic energy"^[5].
• Claim: In running, efficiency increases steadily with speed from ~45% at 2.77 m/s to ~60% at 5.55 m/s^[6].
  • I'm not sure how Garmin came to this conclusion given this research study, but I may have missed something. I could find no mention of measuring oxygen consumption or metabolic work. The published paper does have graphs on page 416 that show the total mechanical work, along with the internal work (used to move your limbs forward into position) and the external work (used to propel the body forward.) It looks to me like Garmin may have used the ratio between the external mechanical work and the total mechanical work to define "efficiency." This is a totally different to the previous study which is looking at the ratio of total energy to mechanical energy, whereas this study is looking at a subdivision of the mechanical energy.
  • This study does raise the interesting question of what should be counted as "running power." Should the power going into your arm swing be considered part of running power or just wasted energy? Likewise, the power used to move your leg forward in the swing phase could be considered wasted energy or running power.
• Claim: "A review paper on cycling efficiencies indicates efficiency for cycling is between 20 to 25% for power between 200-300 W"^[7].
  • The paper says 18.5 to 23.5% at 300W, but this paper actually references another study^[8] that shows cycling gross efficiency of between 18.3 and 22.6%. So the numbers are little different, but close enough.
• Claim: "Some researchers study delta efficiency, or apparent efficiency, which is the ratio of an increment in external mechanical power output to the increase in metabolic power required to produce it."
  • Sub-claim "delta efficiency for running is 45.5%, cycling is 25.7%"^[9].
    • This study actually only considers the mechanical cost of moving uphill, using a simple calculation of mass, gravity, velocity, and the angle of incline. So this isn't the delta efficiency of running, so much as the efficiency of moving the body upwards along an incline.
  • Sub-claim "delta efficiency for running is 45.5%, cycling is 25.7%"^[10].
    • This is a similar study by the same researchers, using the same simplistic measure of efficiency.
  • Sub-claim "delta efficiency for running is 53.8%, cycling is 25.1%"^[11].
    • This study is comparing the changes in efficiency between running and cycling when there is a force pulling the subject backwards. For runners and they used a rope going over a pulley and attached to a weight. From what I can tell, the delta efficiency only looks at the work performed from the distance covered and the force of the weight.
• Claim: "The table below shows external power experienced at the road and does not include internal power required for repositioning the limbs"^[12][13][14]
  • Garmin is correct in saying these studies only look at the external power, not the internal power, but more importantly they ignore the elastic properties of the human body.

None of this research appears to be relevant to real-world running, or helpful in calculating appropriate running power.

9 References

1. ↑ https://support.garmin.com/faqSearch/en-US/faq/content/vAzl2ne5dV6diuc2XbYZA6, Garmin Running Power Higher Compared to Cycling Power, Accessed on 2017-12-30
2. ↑ D. J. Farris, G. S. Sawicki, The mechanics and energetics of human walking and running: a joint level perspective, Journal of The Royal Society Interface, volume 9, issue 66, 2011, pages 110–118, ISSN 1742-5689, doi 10.1098/rsif.2011.0182
3. ↑ T. Fukunaga, K. Kubo, Y. Kawakami, S. Fukashiro, H. Kanehisa, C. N. Maganaris, In vivo behaviour of human muscle tendon during walking, Proceedings of the Royal Society B: Biological Sciences, volume 268, issue 1464, 2001, pages 229–233, ISSN 0962-8452, doi 10.1098/rspb.2000.1361
4. ↑ M. Ishikawa, J. Pakaslahti, P.V. Komi, Medial gastrocnemius muscle behavior during human running and walking, Gait & Posture, volume 25, issue 3, 2007, pages 380–384, ISSN 09666362, doi 10.1016/j.gaitpost.2006.05.002
5. ↑ A. L. Hof, J. P. Van Zandwijk, M. F. Bobbert, Mechanics of human triceps surae muscle in walking, running and jumping, Acta Physiologica Scandinavica, volume 174, issue 1, 2002, pages 17–30, ISSN 0001-6772, doi 10.1046/j.1365-201x.2002.00917.x
6. ↑ G.A Cavagna, M.A Legramandi, L.A Peyre-Tartaruga, Old men running: mechanical work and elastic bounce, Proceedings of the Royal Society B: Biological Sciences, volume 275, issue 1633, 2008, pages 411–418, ISSN 0962-8452, doi 10.1098/rspb.2007.1288
7. ↑ Michael J. Joyner, Edward F. Coyle, Endurance exercise performance: the physiology of champions, The Journal of Physiology, volume 586, issue 1, 2008, pages 35–44, ISSN 00223751, doi 10.1113/jphysiol.2007.143834
8. ↑ Edward F. Coyle, Labros S. Sidossis, Jeffrey F. Horowitz, John D. Beltz, Cycling efficiency is related to the percentage of Type I muscle fibers, Medicine & Science in Sports & Exercise, volume 24, issue 7, 1992, pages 782???788, ISSN 0195-9131, doi 10.1249/00005768-199207000-00008
9. ↑ Kirsten E. Bijker, Gert De Groot, A. Peter Hollander, Delta efficiencies of running and cycling, Medicine & Science in Sports & Exercise, volume 33, issue 9, 2001, pages 1546–1551, ISSN 0195-9131, doi 10.1097/00005768-200109000-00019
10. ↑ Bijker K., G. de Groot, Hollander A., Differences in leg muscle activity during running and cycling in humans, European Journal of Applied Physiology, volume 87, issue 6, 2002, pages 556–561, ISSN 1439-6319, doi 10.1007/s00421-002-0663-8
11. ↑ Erling Asmussen, Flemming Bonde-Petersen, Apparent Efficiency and Storage of Elastic Energy in Human Muscles during Exercise, Acta Physiologica Scandinavica, volume 92, issue 4, 1974, pages 537–545, ISSN 00016772, doi 10.1111/j.1748-1716.1974.tb05776.x
12. ↑ G. A. Cavagna, M. Kaneko, Mechanical work and efficiency in level walking and running, The Journal of Physiology, volume 268, issue 2, 1977, pages 467–481, ISSN 00223751, doi 10.1113/jphysiol.1977.sp011866
13. ↑ Keith R. Williams, Peter R. Cavanagh, A model for the calculation of mechanical power during distance running, Journal of Biomechanics, volume 16, issue 2, 1983, pages 115–128, ISSN 00219290, doi 10.1016/0021-9290(83)90035-0
14. ↑ G. A. Cavagna, M. Mantovani, P. A. Willems, G. Musch, The resonant step frequency in human running, Pflgers Archiv European Journal of Physiology, volume 434, issue 6, 1997, pages 678–684, ISSN 0031-6768, doi 10.1007/s004240050451