Clustering of foot strike patterns when running

While it is well known that the foot strike pattern does vary with velocity, it is not the sole determinant of the foot strike pattern as many elite runners heel strike when running fast and a runner’s self determination of their foot strike pattern appears to be poor. However, the data that quantifies velocity and foot strike pattern is somewhat limited. It is probably easy to discreetly divide the foot strike pattern into forefoot, ‘midfoot’ or heel strike as it is visually reasonably obvious. But not all foot strikes in each pattern are the same. The heel strike may be ‘hard’ or ‘light’ and a heel strike may have a high or low touch down angle. Rather than just classify a pattern of anything based on a conventional classification, dogma or collective wisdom, an approach loosely known as ‘clustering’ is used. In this approach a whole lot of data is collected and then a more sophisticated statistical modelling approach is used to divide the data into distinct clusters to come up with the classification. That classification can then be used for further research. This is what the following study did:

The Effect of Running Velocity on Footstrike Angle–A Curve-Clustering Approach
S.E. Forrester, , J. Townend
Gait & Posture; Available online 18 August 2014
Despite a large number of studies that have considered footstrike pattern, relatively little is known about how runners alter their footstrike pattern with running velocity. The purpose of this study was to determine how footstrike pattern, defined by footstrike angle (FSA), is affected by running velocity in recreational athletes. One hundred and two recreational athletes ran on a treadmill at up to ten set velocities ranging from 2.2–6.1 m·s−1. Footstrike angle (positive rearfoot strike, negative forefoot strike), as well as stride frequency, normalised stride length, ground contact time and duty factor, were obtained from sagittal plane high speed video captured at 240 Hz. A probabilistic curve-clustering method was applied to the FSA data of all participants. The curve-clustering analysis identified three distinct and approximately equally sized groups of behaviour: (1) small/negative FSA throughout; (2) large positive FSA at low velocities (≤4 m·s−1) transitioning to a smaller FSA at higher velocities (≥5 m·s−1); (3) large positive FSA throughout. As expected, stride frequency was higher, while normalised stride length, ground contact time and duty factor were all lower for Cluster 1 compared to Cluster 3 across all velocities; Cluster 2 typically displayed intermediate values. These three clusters of FSA–velocity behaviour, and in particular the two differing trends observed in runners with a large positive FSAs at lower velocities, can provide a novel and relevant means of grouping athletes for further assessment of their running biomechanics

I won’t try to explain the details as I will be here all day¹. Nothing jumps out at me as being problematic. They did use a treadmill, so some may have an issue with that (I don’t, due to the nature of the research question). The authors did identify a limitation of the fixed velocities rather than self selected velocities, but I do not have a problem with that.

Using the discrete foot strike classification they found:

  • for velocities between 2.2-4.9ms-1, about 70% were rearfoot strikers; 24% were midfoot and 6% forefoot
  • at 6.1 ms-1, about 55% were rearfoot strikers; 38% were midfoot and 7% forefoot
  • so with increased velocity there was a trend towards midfoot away from heel strike

However, using the curve cluster analysis, there were 3 distinct clusters which coincidentally ended up with  almost the same number in each group:

  • Cluster 1: small or negative foot strike angle through all velocities
  • Cluster 2: large positive foot strike angle at low velocities that transitioned to a smaller foot strike angle at higher velocities
  • Cluster 3: large positive foot strike angle throughout all velocities

Here is the pictorial representation from the paper:
cluster

So, the foot strike angle is affected by velocity in some recreational runners and not others (ie its subject specific!). The study did have some other findings, but this was the key one for me. It should be obvious to see the implications of this. What if the previous research that looked at injury rates or running economy for different foot strike patterns were redone using the above clusters? Will the results be the same or different?

As always, I go where the evidence takes me until convinced otherwise ….. and this is a novel approach to classifying foot strike pattern.

1. Its fathers day here in Australia.

Forrester, S., & Townend, J. (2014). The Effect of Running Velocity on Footstrike Angle–A Curve-Clustering Approach Gait & Posture DOI: 10.1016/j.gaitpost.2014.08.004

Last updated by .

,

3 Responses to Clustering of foot strike patterns when running

  1. Daniel Petcu September 8, 2014 at 5:02 pm #

    Dear Dr. Payne,

    I think another one interesting approach of using clustering in running could be found in:
    Steffen Willwacher , Katina Fischer , Joseph Hamill , Eric Rohr & Peter Brueggemann (2013) Free moment patterns in distance running, Footwear Science, 5:sup1, S10-S11, DOI: 10.1080/19424280.2013.798686

    Respectfully,
    Daniel

  2. eric johnson September 9, 2014 at 3:50 pm #

    hey craig,

    seems like this is one major factor that explains why it’s hard to correlate foot strike with running speed. would you agree or do you think there are other factors at play?

    thanks.

    • Craig Payne September 9, 2014 at 7:03 pm #

      There are many many factors at play. I increasingly believing that foot strike pattern is not that important (with the exception of ‘overstriding’) – take care of all those other factors and the foot falls where it falls.

Leave a Reply