The impact of height is a great point. Ryan and Gebre are taller marathon runners as compared to most of the 5 ft. 2in. to 5ft. 4 in. runners so they don’t necessarily need as high a cadence and in fact it may be inefficient to do so at that height. I don’t know how tall Kilel is but the commentator kept referring to her long legs during the final stretch of the race. As a 6ft 1in. runner, I find it nearly impossible to learn all that much from much shorter runners. I tend to look at runner more around my height.

Lastly, Pete, based on Ryan’s data (at this point in time) would you say he has more vertical movement than the others?

Of course, the sample is way too small, but if I were forced at gunpoint to draw a conclusion from the graph, it would be that the variability in swingtime and steptime in those with a high step rate is larger.

Mosop and Kilel have nearly equal step times*. If you draw an imaginary line between the points representing Mosop and Kilel, those under it have the highest step rate and those above it have a low step rate. The further they are from this imaginary line, the higher/lower their step rate.

The ‘cloud’ of the four athleteswith the highest step rate -those under or on the imaginary line- (i.e. Mosop Davila Mutai Goucher) is much more spread out (perpendicular to the line between Mosop and Kilel) than the cloud of athletes with a low step rate.

I’d be interested to see if this holds if mere mortals are added.

BTW love your blog and (as a result) bought me a pair of Kinvaras.

*)and step rates. There seems to be a slight error in the step rate of Kilel. 60/0.322 = 166.3

Would be interesting – we’d need high speed video to figure it out. Then you
add in variability due to course, location in race, etc. Makes me want to go
through my Boston videos from last year and do a detailed anaylsis. Would
take a long time, but would be interesting.

Rinie,

The error was due to the way I calculated step rate – it was from the mean
of step rates calculated independently for each foot, and not by dividing 60
by the mean step time. Thus, the discrepancy I think is because of rounding
error and the number of digits past the decimal used in the calculations. I
believe your correction should read 186.3, so not a big difference anyway.

Pete

All of these suggestions revolve around trying to normalize the stats so that it’s we can better compare running gaits. For instance the tall runners will need a lower cadence than a short runner for the same running gait, that is, if one assumes the angles of the various body parts through the cycle being similar being a marker of similar running gait.

“Lastly, Pete, based on Ryan’s data (at this point in time) would you say he has more vertical movement than the others?”

The best way to derive this from the footage is to use the time in the air and the a mathematical model of parabolic flight under gravity. I’ve done some quick maths, assuming that the toe off height and landing height are the same, the equation would be g * T * T / 8.

Admitedly not a perfect model given toe off and landing are likely to be a little different height, but it should be good enough to understand the basic relationsip – double the time in the air and you quadruple the height, this make sense as you have to launch twice as fast vertically, and it takes twice as long to reach the peak, multiple the two together and you get four times the height.

One can simplify it further and just say, the longer in the air, the higher the peak height. And for these runners Mosop has the longest time in the air and with it greatest vertical height achieved, and has 32% longer time in the air than Mutai, and with it a 74% higher maximum height. With the square law the height estimates will be very sensitive to inaccuracies with Pete’s time estimates, but the overall relationships are still solid. One thing I’d take away is that measuring height oscilation clearly isn’t a great indicator of performance – we have a huge variation here, but little actual performance difference or absolutely no consisent for this small dataset.

One should also be able to estimate the average loading on stance using the time in air vs time on stance. I’ll leave the calcs for another day though.