The Training Intensity Distribution of Marathon Runners Across Performance Levels
Protocol
n=119452 runners (119,452) who completed 151,813 marathons; training sessions from the 16 weeks preceding each marathon were analysed (Strava data)
TID quantified with a three-zone approach: critical speed defines the Z2/Z3 boundary; the Z1/Z2 transition was assumed at 82.3% of critical speed
Training characteristics and TID reported by marathon finish time
Key Results
Training volume across all runners: 45.1 ± 26.4 km/week; the fastest runners (marathon 120-150 min) accumulated more than three times the volume of slower runners
Time in Z2 and Z3 stayed relatively stable across performance levels; the proportion of Z1 was higher in progressively faster groups
The most common TID was pyramidal, adopted by more than 80% of runners with the fastest marathon times
Strong negative correlations (p < 0.01, R² ≥ 0.90) between marathon time and markers of training volume and the proportion of volume in Z1 — more volume and a larger Z1 share went with faster times
The proportions of training in Z2 and Z3 were correlated (p < 0.01, R² ≥ 0.85) with slower marathon times
Conclusion: the fastest runners featured large volumes achieved primarily by adding Z1; marathoners adopted a pyramidal TID and its prevalence increased in the fastest runners
Practical Application
Across 119,452 marathon runners (151,813 marathons, 16 weeks of Strava logs each), faster finishers trained more — the 120-150 min group did over three times the volume of slower runners (overall mean 45.1 km/week) — and the extra volume was almost entirely Z1 (below 82.3% of critical speed): time in Z2 and Z3 was similar across levels, the Z1 share rose with speed (R² ≥ 0.90 against finish time), and larger Z2/Z3 shares went with slower times (R² ≥ 0.85).The dominant distribution at every level was pyramidal, in more than 80% of the fastest runners — not polarised.
What separates faster from slower marathoners in this dataset is the size of the easy layer, not a bigger dose of hard running; time above critical speed stayed roughly constant while the base grew around it.
Inference, not tested: this is cross-sectional — faster runners may tolerate more Z1 because they are faster; it does not prove that adding Z1 makes a given runner faster. It does show that a larger Z2/Z3 share is not what fast marathoners do.
Mechanisms
Inference, not tested: a stable absolute amount of Z2/Z3 with a growing Z1 layer is what produces the pyramidal shape; the abstract offers no mechanism and the design cannot test one.
Limitations
Cross-sectional: association, not causation; faster runners may be those who can tolerate more volume
Zones estimated from Strava data; the Z1/Z2 boundary was assumed at 82.3% of critical speed rather than measured
Running only (marathon); transfer to cycling is discussion, not data
Self-selected Strava users; no physiological testing
How this entry was checked and changed
Why this level
regraded 2026-09-27 from A: observational analysis of training logs, no intervention
Content correction 2026-09-27
"n=180 (45 elite / 90 competitive / 45 recreational), 12-month logs, VO2max ~70 / 55-60 / <50" -> 119,452 runners, 151,813 marathons, 16 weeks of Strava logs before each; no VO2max data.
Polarised table "78/8/14, 72/18/10, 65/25/10" and "elite use more polarised TID, supporting Seiler" -> deleted; the abstract reports a pyramidal TID in more than 80% of the fastest runners, stable Z2/Z3 time across levels and a Z1 share that rises with speed (R² ≥ 0.90).
Title suffix "Validation of Polarized Model in Endurance Running" -> removed (not the paper's title or finding). "Across endurance sports (running, cycling)... 70-80 / 5-10 / 15-20" -> deleted; running only.
Sport line cycling, running -> running. Level A -> B+. Cross-reference "REF-143 (Seiler 2010)" -> REF-087 (Seiler 2010); REF-143 is Gallo 2024.
Cross-references to duplicate stubs repointed to their canonical entries (REF-085->REF-303, REF-162->REF-306); the stubs were emptied on 2026-09-27.
Cross-reference descriptions checked against the cited entries' current content; wrong years, wrong entries and claims the cited entry does not make were corrected.
2026-10-01: cross-reference repointed from a merged framework to the one that replaced it (SYNTH-01→31, 05→34, 09→70, 12/15→32, 19→26).
Full text (checked 2026-10-12)
The dataset held 28,118 female and 91,334 male runners whose marathons were run between 2014 and 2017; mean finish time was 230.2 ± 41.9 min (full text, Table 1).
Runners averaged 3.6 ± 1.7 sessions per week, and the average weekly long run was 19.6 ± 9.6 km (full text, Table 1).
The 120-150 min finishers ran ~107 km·week-1 (n = 620), against ~35 km·week-1 for runners finishing in 270-300 min (n = 8,798) (full text).
Across all runners, 49.0% of training was in Z1, 35.3% in Z2 and 15.7% in Z3; the polarisation index was 1.25 overall and stayed below 1.5 at every finish time (full text).
The highest Z1 share was in the fastest males (~67%) and females (~57%), below the ~80% typically reported for elite athletes (full text).
Pace was grade-adjusted (Minetti method), and critical speed set the Z2/Z3 boundary (full text).
Regression models of finish time using the Z1, Z2 or Z3 fraction plus volume, long runs, polarisation index and sex had adjusted R2 of 0.596, 0.592 and 0.600 respectively (full text, Table 2).