Running the Numbers: Data on my First Marathon

I recently completed the Marine Corps Marathon in DC! This was my first marathon and I wanted to share some of the data I collected along the way, both in training and from the actual race.

Results

My pace was solid for the first half marathon, but I started to hit a wall. I think I came off a little too strong and ran out of gas towards the end. I maintained a roughly 8.5 minute pace for the first half, and ended with over an 11 minute pace.

Running the Numbers

I wanted to see how my own results compared to all the other runners. I was particularly interested in the shape of the distribution of finishing times and the graph of pace over distance for different finishing times. I pulled all the data from the MCM website (more details on how I did this below) and here's what I found:

Distribution of Finishing Times

The finishing times were a bit more normally distributed than I expected. There was a cluster of finishers between 3:45 and 4:15, but other than that the distribution was pretty evenly spread out.

Race Splits for Different Pace Zones

In the chart below, I grabbed a random candidate from each of the 30 minute finishing time buckets from under 2:30 to under 7:30. I plotted the pace over distance for each candidate to see how pace changed throughout the race for runners with different speed profiles.

Unsuprisingly, the fastest runners had a very consistent pace, while the slower runners had much more variance in their pace profiles over the course of the race. The better performing runners also tended to speed up a bit after mile 20, while the slower runners tended to consistently slow down after mile 10.

Training

I started tracking my mileage 18 weeks out from the race. I did not follow a strict training schedule, but focused on incremental and consistent mileage. My strategy was to maintain 10-20 miles a week. I would typically do a long run on Saturday or Sunday, and one or two shorter to medium distance runs during the week. I gradually increased my long run from 10 miles to a maximum of 19. In terms of intensity I mixed in some shorter faster paced runs along with my slower paced long runs. I was able to stay pretty consistent with this schedule, exluding a couple of weeks in early October where I was sick and had a nagging injury.

Next Time

Ultimately, I really enjoyed my first marathon, and I plan on doing another one. I've taken a break from running the last few weeks, but am planning to get back into it soon. For next time, I think I would adjust my training schedule to get some more consistent mileage. I think having some more milage under my feet would have helped me hold on to a stronger pace in the second half of the race. Whenever I do decide to do my next marathon, I'll be sure to collect more data as I go.

Appendix: Gathering the Data

I'm a big data and numbers guy (if you can't tell), and I always really enjoy figuring out how to extract data like this into a useful format. In this case I was able to just use the public API used by the MCM website.

Extracting the Data

• https://api.rtrt.me/events/MCM-2024/profiles
• https://api.rtrt.me/events/MCM-2024/profiles/{pid}/splits

Processing Fast with Threads

Once I got all the code good to go to extract and parse the data for a single bib number, I setup a thread pool so that I could collect all the data async pretty quickly. I experimented with a few different thread counts to get a feel for the throughput, error rates, and any potential rate limiting. I started off with a chunk of 100 bibs:

I then increased to a chunk of 1000.

I found that the number of errors was consistent across the thread counts, so I thought it was safe to assume I wasn't being rate limited. I also found that the marginal improvement on throughput on my local machine started to drop off after ~100 threads. I ended up going with 500 threads for the full extraction.

Within 6 minutes, I was able to extract all the data to a ~50mb json file on my local machine.