A Data-Driven Approach to Better Understand Shoe Life

Modern footwear is made from high quality materials, but no matter how durable shoe soles are, all materials degrade over time. So how long should we wear running shoes before they go from offering great shock absorbing capability to posing more of a risk for running injury? Is there a way to better understand the degradation of footwear material and also measure it?

Footwear offers the most cushioning right out of the box but decreases with more wear, posing a risk for greater stresses and force through the body upon foot strike.1,2,3,9 There are several ways in which materials degrade but even the rate of degradation is assumed to be influenced by a plethora of factors. The most well-known factor is mileage. Several studies have shown that as the number of miles ran in a shoe increases, the shock absorbing capability in the midsole decrease.1,3,4,9 Mileage is easy to keep track of and Under Armour Connected Footwear does exactly that, but as research has shown, mileage is just one of many factors that can affect material degradation.

Every runner is unique and individual characteristics such as body mass, footwear type, and running pattern affect shoe wear differently.1,4,5,6,7,8,9 It is hypothesized in the running community that a heavier runner will put more material stress on their shoes than a lighter one.5,6,7,8 Another study showed that even the type of footwear should be considered when looking at the longevity of cushioning capacity. They compared a minimalistic vs traditional shoe after undergoing the same amount of mechanical aging (a process that simulates real life foot impact after impact to wear out the material) and found that thicker and softer heel foams absorbed 83% more energy but degraded at a 49% faster rate.4 Under Armour currently recommends using a pair of shoes for 350 miles, but what other factors can our products track to create a more personalized approach to understanding how long running shoes should last? Simply setting a cap to 350 miles is a one-size-fits-all approach to something that is intuitively more personal. 

To further investigate the effect of body weight and other individual factors such as average pace during a workout, height and weight, we partnered with the Under Armour data science team to better characterize use patterns of our own MapMyFitness community. We hypothesized that Connected Footwear consumers with a higher body mass will retire their shoes before consumers with a lower body mass. 

Data Experiment 

To test this hypothesis, we identified footwear that had been used extensively and were no longer in use. We deemed this footwear “retired”. To define “retired” we came up with a set of criteria like a minimum threshold for miles, certain period of inactivity while user is still active, etc. Distributions of a few variables from this dataset are shown in Figure 1. Total distance on the shoes refers to the distance on shoes before they were retired. The rest of the variables relate to consumers’ characteristics.

Figure 1: Distribution of shoes and users’ characteristics

To explore how individuals physical characteristics and running habits affect the number of miles logged on a pair of shoes before retirement, Pearson correlation coefficients were computed. To conduct a regression analysis, we needed to define target and predictor variables. A target variable is the output which needs to be predicted, while a predictor is an input variable which we hope to predict the target. The target variable was total distance on retired shoes while predictor variables were user characteristics such as age, average workout pace, height, weight and their combination BMI. During the exploratory data analysis phase, the association of each independent predictor with the target variable was studied. We discovered the Pearson correlation coefficient between the target and each of the predictors is close to zero, indicating an almost non-existent linear correlation between the pair of variables. Figure 2 shows how patternless the relationship between users’ weight and total distance on their shoes can be. 

Figure 2: Consumers’ weight vs. total distance on their shoes

Next, we used a linear regression approach to test whether a combination of predictors together would predict the total distance on the shoes. This effort was unsuccessful resulting in high prediction errors. 

Conclusion

Leveraging a pure machine learning approach to verify the hypothesis suggested the hypothesis was null. However, when working with the data set we concluded a few things. First, we had assumed the main reason retired shoes were out of use was due to being worn out. We concluded this to be a weak assumption since users may have stopped using the shoes for a variety of other reasons such as injury from the shoes, or losing interest in them. We also found that consumers stopped using their shoes at the 350 miles threshold which is recommended by UA. Lastly, our dataset may have had bias due to a small sample size of consumers with a high body mass.

The characteristics of the studied data set made it difficult to test the hypothesis fairly. For future research we may consider running a controlled longitudinal experiment in which active athletes with different body masses and their shoe life are monitored by an expert. In summary, a different approach to study why consumers retire or stop using their footwear would deem useful to better capture use patterns.

References

  1. Chambon, Nicolas, et al. “Aging of Running Shoes and Its Effect on Mechanical and Biomechanical Variables: Implications for Runners.” Journal of Sports Sciences, vol. 32, no. 11, 2014, pp. 1013–1022.
  2. Goonetilleke, Ravindra S. “Footwear Cushioning: Relating Objective and Subjective Measurements.” Human Factors: The Journal of the Human Factors and Ergonomics Society, vol. 41, no. 2, 1999, pp. 241–256.
  3. Heidenfelder, Jens, et al. “Biomechanical Wear Testing of Running Shoes.” Footwear Science, vol. 1, no. sup1, 2009, pp. 16–17.
  4. Lippa, Nadine M., et al. “Mechanical Aging Performance of Minimalist and Traditional Footwear Foams.” Footwear Science, vol. 9, no. 1, 2016, pp. 9–20.
  5. Michel, Katja J, et al. “The Effect of Different Midsole Hardness on Kinematic and Kinetic Data during Running Influence by Varying Bodyweight.”
  6. Runner’s World. “Weighty Matters: Running Shoes & Body Weight.” Runner’s World, Runner’s World, 17 Jan. 2019, www.runnersworld.com/uk/gear/a764281/weighty-matters-running-shoes-amp-body-weight/.
  7. Runtastic. “How To Choose the Best Running Shoes >> 6 Factors.” Runtastic Blog, Runtastic, 4 Mar. 2019, www.runtastic.com/blog/en/running-shoes/.
  8. UniSA. “Heavy Runners Risk Injury in Lightweight Running Shoes.” UniSA News Releases – University of South Australia, University of South Australia, 1 Nov. 2018, www.unisa.edu.au/Media-Centre/Releases/2017-Media-Releases/Heavy-runners-risk-injury-in-lightweight-running-shoes/#.WUvzcRPyvBK.
  9. Wang, Lin, et al. “Changes in Heel Cushioning Characteristics of Running Shoes with Running Mileage.” Footwear Science, vol. 2, no. 3, 2010, pp. 141–147.
connected footweardatashoe life