Come 26 October, roughly 20,000 runners will line Leeson Street ready to take on the Dublin Marathon. For approximately 30%, it will be their first-ever Dublin marathon.
Many will hold out hopes for a shiny new personal best. Others will be dealing with gnawing anxieties, contemplating the months of training and questioning whether it was sufficient, second-guessing their pacing strategy and dwelling on fueling plans.
When it comes to running a 'good' marathon, what does the data have to say?
Barry Smyth, a professor of computer science at UCD, has spent a career combining his expertise in data science with his passion for marathon running, extracting the key lessons in tackling 42.195 kilometres.
Smyth has sifted through thousands of marathon runners’ profiles on Strava (a world-popular digital training log app), and mapped the data with AI. He highlighted the key learnings and potholes to avoid while speaking at a Talk of the Town event in Dublin.
The '80/20 rule' has informed the training of many of the world’s best marathoners for decades. The idea states that the majority of training (80%) should take place at a comfortable, conversational pace while the remainder is reserved for high-quality, intense workouts that drive physiological adaptation.
The data validates the philosophy. Elite/sub-elite athletes who finish the marathon in 2:00 to 2:30 hours will generally spend over 70% of their time in moderate heart zones (Z1, as indicated, in the graphic below).
As finishing times slow, the less time an athlete is likely to have spent in these easy zones. For those finishing between 4:00 and 4:30 hours, their training represents an even split between comfortable running and more intense training (Z2/Z3).
Another underpinning principle of many marathon training blocks is the ‘10% rule’, which stresses mileage should never increase by more than 10% week on week in order to avoid injuries.
While Smyth says the 10% rule is outdated, acute chronic workload ratio (ACWR), a comparison between an athlete’s latest weekly mileage and the average of their previous four weeks, functions as a far better barometer.
If an athlete had run a weekly average of 50 kilometres for the past four weeks, before then running 55 kilometres the next week, their ACWR would be 1.1.
Data from over 30,000 runners shows athletes who maintained a 1.1 ratio were far less likely to experience substantial training breaks. An athlete who stuck around the 1.1 ratio would typically miss five days of training, while runners north of 2.0 averaged nine days out of action.
The decision on when to lower training volume and begin ‘tapering’ is another variable that haunts marathoners.
An effective taper primes the body for race day, but purposefully training less can quell anxiety over fears of losing fitness.
Smyth’s analysis of 150,000 marathoners’ last four weeks before race day versus weeks five and six pre-race (typically peak training volume) shows most began tapering two to three weeks out.
Discipline proves key. While some runners implemented a ‘relaxed’ taper where their weekly mileage decreased on average but fluctuated up and down week by week, a linear downshift yielded far better results.
Those who carried out a ‘strict’ taper shaved an average of a minute and a half off their times compared to their more relaxed counterparts.
When all the training is done, Smyth emphasises the importance of keeping the race-day adrenaline at bay to ensure previous months of gruelling training aren’t wasted on a gung-ho pacing strategy.
Data shows 70% of runners’ first five kilometres are faster than their average marathon pace, with the challenge lying in racing steadily while still expending all energy systems by the finish line.
Smyth advocates for the ‘controlled fade’, which involves a 0.5 to 3% downturn in pace (positive split) from the first half of the marathon to the second.
Another interesting finding? The stereotype of men being egotistical loose cannons powered by testosterone and bravado may not be a grand stretch.
Men are almost 65% more likely than women to hit the wall (experience a 25%+ downturn in pace during the second half of the race), and more likely to record large positive splits.
The worlds of running, AI and big data continue to become more intertwined.
Many who toe the line of the Dublin Marathon will have used Runna, an AI training platform now owned by Strava, which also offers AI-backed estimates of athletes’ fitness levels and likelihood of overtraining.
"(Strava and Runna) are using a lot of existing machine learning techniques. The reason they're able to do it so well is they've got so much data," Smyth told RTÉ Sport.
He believes artificially generated training plans will continue growing in lockstep with AI advancement.
"If you look at a typical paper training plan, it'll tell you to run this session at your lactate threshold; (people) won't know what that is, and Runna will at least have an estimate of that," he said.
"I think just helping people to understand what the right pace to run these sessions and to make sure they're doing that pace not too fast… that speed is chosen because it's at a particular physiological point, and if you go too fast, you might get a little bit benefit, but not as much, or you're at the point of diminishing returns.
"I think those apps that are actively monitoring you and tweaking the training plans based on what they see, they have the potential to be better than a fixed, static training plan."
