What’s the Most Reliable Predictor of Your Marathon Time?

A few years ago, I wrote an article about a high-tech marathon prediction study that crunched Strava data from 25,000 runners. They extracted each runner’s fastest training segments over distances ranging from 400 meters to 5K, plotted the data as a hyperbolic speed-versus-duration curve, used that curve to calculate the runner’s critical speed, and used the critical speed to predict their marathon time.

If none of that made sense to you, or if you don’t have a GPS watch, or if you simply can’t be bothered to upload all your training data into an all-seeing algorithm, then I’ve got a different kind of marathon prediction study for you. In the European Journal of Applied Physiology, Japanese researchers led by Akihiko Yamaguchi look at simpler variables like how much and how often you run, and come up with some big-picture insights that are worth bearing in mind next time you tackle 26.2 miles.

The researchers surveyed about 500 runners about their training habits leading up to the Hokkaido Marathon, focusing on monthly training volume, number of running days per week, average run distance, and longest run distance. (According to the paper, Japanese runners and running media generally track their training volume by month, rather than the weekly totals more common in North America.)

Astute readers will notice that these variables are interconnected: if you know running frequency and average run distance, then you’ve already specified monthly training volume. That’s what makes this sort of analysis tricky. Lots of previous studies have tried to figure out which training variables are the best predictor of marathon time. But if, say, total training volume is a good predictor, it’s hard to know whether that’s because running every day is the most important thing, or whether having some really long runs is the key, or whether total mileage is what matters, regardless of how you accrue it.

To get around this, the researchers divided their runners into subgroups. For example, they created four subgroups of monthly mileage: those who ran less than 100K (62 miles) per month; 101 to 150K; 151 to 200K; and over 200K. Within each of those groups, monthly mileage had no power to predict who would run the fastest marathon, because everyone was doing similar mileage. Then you can ask what variables do predict marathon time. Is it running frequency? Average run distance? Longest run distance? The answer, interestingly, is that none of them have any significant predictive power. For people running similar overall mileage, the other training variables tell you nothing useful.

They followed a similar procedure for training frequency, dividing the subjects into homogeneous groups running one to two times a week, three to four times, and five to seven times, then analyzing the effect of the other variables. In this case, the strongest predictor was monthly mileage: for a given running frequency, the more you run, the better. Average run distance was also a predictor, but that’s not adding anything new: if you’re running the same number of days per week, then those with higher average run distance will also have higher monthly mileage.

Subgrouping the other two variables (average run distance and longest run distance) produced similar results: in each case, total monthly mileage was the best predictor of marathon time within each subgroup. But that relationship only held for people whose average run was at least six miles and whose longest run was at least 12 miles. Below a certain minimum level of training, all predictions are off.

So far, this might seem painfully obvious: those who run more mileage race faster marathons. But the subgroup analysis allows us to draw some stronger conclusions. Most notably, it doesn’t seem to matter how you accumulate that mileage: a bunch of short runs or a few long runs produce similar results. That parallels findings from earlier this summer in JAMA Internal Medicine about the health benefits of being a so-called weekend warrior: long-term mortality depends on how much exercise you get, but it doesn’t matter whether you spread your exercise throughout the week or pack it in on the weekend.

If you dig further into the subgroup analyses, you also find that the longest run was a better predictor than the average run. As a result, the researchers conclude that at a given level of mileage, it’s better to do one long run and several short ones rather than doing all your runs at a similar distance. This, too, lines up with marathon orthodoxy that says there’s no substitute for long runs.

Compared to the 25,000-runner Strava study, this one has a lot of shortcomings. It’s very small, the training data is self-reported and (as a result) doesn’t include any measure of speed, the subjects are very lightly trained (averaging 93 miles per month, or roughly 23 miles per week, with an average finishing time of 4:20). If you’re looking to qualify for the next Olympics, or even for Boston, don’t look for any secrets here: you should be accumulating volume and frequency and long runs, not trying to figure out which variables you can neglect.

But there are times in the life of every runner when training slips down a few notches on your priority list. In those situations, the rule of thumb from this study seems more useful than the formula for how to calculate critical speed from your Strava data. The rule is: accumulate as much mileage as you can, whenever you can, in whatever dose you can get it. Sometimes the runs may be shorter or less frequent than you like, but come race day it all counts.

For more Sweat Science, join me on Twitter and Facebook, sign up for the email newsletter, and check out my book Endure: Mind, Body, and the Curiously Elastic Limits of Human Performance.