Predicting Marathon Time
Approach
My formula for predicting marathon time is based on Pete Riegel’s formula that was published in Runner’s World in August of 1977. Riegel was a research engineer as well as a marathon runner, and his formula has been used by runners around the world for the past 20 years.
The formula:
T2 = T1 x (D2/D1) 1.06
Where:
T1 = Given Time
D1 = Given Distance
D2 = Distance to predict a time for
T2 = calculated Time for D2
This formula implied that your overall pace will slow down as you run longer distances
Boundaries
Boundary values for the degradation coefficient are 1.06 as the fastest time and 1.10 for the slowest time. Additionally, a score for the elevation and distance is set as 0% for the lowest value and 3% as the highest value
​
​
degradation_coefficient_low = 1.06
degradation_coefficient_high = 1.1
max_score = 3
Why use Riegel over a linear function?
When I tried using the linear formula for my initial analysis, the prediction times were a bit off. While linear does provide boundary values that help for estimation, Riedel’s formula accounts for more factors to get a more accurate representation. You can see a side by side of how the numbers vary in my final output.
Working with Time
My Garmin inputs data in hours, minutes, and seconds, which is difficult to work with when trying to perform an analysis. Due to this, I converted the time data into seconds to be used in algorithms and then eventually turned back into their standard format.
Coefficient Score
RESULTS
The final output prints the linear prediction time as well as the low, medium, and high degradation coefficient rates based on my running data. After examining the data Run Log number 107’s Pete Riegel with a High Gap prediction most accurately matched my Marathon time. I ran the Chicago Marathon in 3 hours and 47 minutes and the predicted time was 3 hours and 48 minutes.
​
Each run log projected slightly different times, some of which were way off, for example Run Log number 100 predicted a Pete Riegel with a high GAP time of 4 hours and 49 minutes. This is because each week varied on time and distance, and without looking further into outside factors such as fatigue, weather, injury, etc. I am not fully able to state why there is such a big difference in times.
My next steps for this project will be adding in my Strava data of other activities completed during the week that could influence performance as well as examining temperature and mileage per week. Overall I am happy with the results from this analysis and am looking forward to further exploration.