DOE-ing Myself: Using design of experiments to run more efficiently
Apr 9, 2020 6:15 AM
| Last Modified: Apr 9, 2020 10:58 AM
So this is me: I am a fairly healthy 36-year-old male living in Austin, Texas, USA. While somewhat active throughout my teens and twenties, post-marriage and kids has seen me become a bit too sedentary, a bit too lazy, and much too easily winded. In May 2019, my wife and I took our kids for a hike, with me bringing our youngest (at the time) in a backpack carrier. After 2+ miles of fairly level terrain, I was exhausted. This precipitated a sea change in my activity.
On 27 May 2019, I started running. It was terrible. I hated it, yet I loved it. Three months later, I was talked into running a 10K trail race (at night), and I was hooked! Four months later, I heroically and quite stupidly ran a 50K trail race. It took me over an hour longer than I’d anticipated. I couldn’t walk for a week, and when I crossed the finish line, it took every ounce of self-control to not openly weep.
Kilometer 27-ish of the Bandera 50K. I completely tanked about 2 kilometers later but still managed to (barely) finish the race.
Fast-forward to today: I am ramping up for a series of four races longer than 60K over the next 10 months. Time to train is at a premium. So as I am running, I am continually asking myself this question: What is the most efficient way to improve my running economy in a minimum amount of time? Enter design of experiments (DOE).
DOE is a mechanism to most efficiently use selected input variables to draw conclusions about pre-determined output variables. If it’s good enough to design top-of-the-line semi-conductor components, it ought to be good enough to help me run faster without running longer. So, what is my plan?
Well, there are some variables that I cannot control but I know are important: weather, mileage (I am using a prescribed training plan), hours of sleep, my weight, etc. These parameters are being tracked, just not controlled. But there are parameters I can control:
Warm up — a 5-minute warmup on a rowing machine (I never warm up.)
Caffeine — drinking coffee right before my run (I prefer my coffee after I run.)
Timing of run — first thing in the morning or afternoon (I love morning runs.)
Shoes — trail or road shoes (Usually it's road shoes on the road and trail shoes on the trail.)
Diet — Did I eat meat in the previous 24 hours? (I’m very curious to see if this has an effect.)
Running hydration — water or electrolyte (I prefer water.)
Apparel — 7”, 5”, or 2” inseam shorts*
*A fellow JMP Systems Engineer (and endurance runner) and I have a pet theory that shorter shorts make you run faster. It’s time to test this theory once and for all.
I will be completing five runs a week for three weeks (see the table below). Through all my runs, I will try to maintain a Zone 2 heart rate to ensure the effort I put into each run is quantitatively consistent. For me, this means I will try to maintain a heart rate between 139 and 144 beats per minute (bpm), while trying to stay as close to 144 bpm as possible. I will also temporarily sacrifice my love of trail running and keep all my runs on roads/sidewalks to maintain consistency.
The JMP-generated DOE trials
So how will I track progress? In addition to recording variables I cannot control, I will specifically track these three parameters:
Grade adjusted pace (GAP) — pace in minutes per mile normalized for topography
Average heart rate — while trying to maintain an average heart rate of 144 bpm, variability is sure to occur
Perceived exertion — a subjective 1-10 measure of how I felt immediately after the run
An improvement in these three outputs is how I am gauging an improvement in efficiency — can I run faster, with a lower heart rate, and upon completing a run, do I feel like I exerted less energy.
Now, if I complete 15 runs over three weeks I will hopefully improve my running efficiency even without controlling these inputs, just by the sheer fact that I am running regularly. To account for this, my design uses week as a blocking variable. As a result, when I build the model upon completing this experiment, JMP will account for week-dependent impacts in running efficiency and subtract them out of the model. This will allow me to better estimate the impact of the factors am trying to study.