The long-running argument about whether recreational marathoners should train polarised — most volume at low intensity, the rest at high — or pyramidal — most at low, with a meaningful middle band of threshold work — has a new and unusually nuanced data point. A study published this month in Scientific Reports, drawing on 16 weeks of tracked training from 120 sub-elite marathon runners, finds that polarised training produced superior marathon performance improvements on average, but that the response is not uniform across the field. Roughly a third of runners do better on a pyramidal plan, another fifth respond well to either, and almost one in five do not improve meaningfully on either approach.
The headline number is the average treatment effect: polarised training improved marathon time by 11.3 ± 3.2 minutes over the 16-week block, against 8.7 ± 2.8 minutes for pyramidal — a 30 percent larger swing despite the polarised group running 12 percent less total weekly volume. The polarised plan placed roughly 80 percent of weekly minutes in Zone 1, 5 percent in Zone 2 and 15 percent in Zone 3; the pyramidal plan, by contrast, placed 75 percent in Zone 1, 20 percent in Zone 2 and 5 percent in Zone 3. Both groups maintained the same long-run distance, race-pace tempo, and weekly hill work.
What is genuinely new is the responder analysis. Using a k-means clustering algorithm on the 16-week longitudinal data, the authors identified four distinct response profiles that do not correlate cleanly with age, weekly volume or starting fitness. "Polarised responders" — 31.5 percent of the cohort — improved meaningfully on polarised but not on pyramidal. "Pyramidal responders" — 31.9 percent — showed the opposite pattern. "Dual responders," 18.7 percent, improved on both, and "non-responders," 17.9 percent, improved on neither. Crucially, no pre-test variable predicted membership of any cluster, including VO₂max, weekly mileage or training age.
The practical implication is awkward for the polarised consensus that has dominated coaching social media for the past five years. The data still favours polarised on average — and for a runner with no prior basis for choosing, polarised is the percentage play — but the existence of a substantial pyramidal-responder cohort means a runner who has tried polarised and stalled is not necessarily failing the workout. They may simply be in a cluster that needs more time at threshold. The authors suggest a six-week trial block as the minimum required to distinguish responders, and stress that switching too early masks the signal.
What none of this resolves is the deeper question — why two seemingly similar runners respond so differently to the same training distribution. The paper points speculatively at fast-twitch fibre composition, mitochondrial enzyme expression and cardiac stroke-volume capacity, but acknowledges these are post-hoc hypotheses that the study was not designed to test. For the working coach, the take-home is more modest: stop assuming polarised is automatically right for every athlete, build in a structured comparison block early in the year, and trust the four-week trend rather than any single workout. The training-intensity-distribution debate is not resolved; it is, more honestly, decentralised.