Understanding sustained slowdown.

Explore one form of late-race slowing, recalculated from the same data release as the rest of the study.

The study

How often do marathon runners experience sustained slowdown?

30.8% of eligible finishes meet the sustained slowdown definition.

1,023,450 of 3,328,159 eligible finishes. The same runner may appear more than once.

3,328,159 eligible finishes

How we measured it. At least 25% slower than the 3.11–12.43 mi reference pace for contiguous recorded sections totaling at least 3.11 mi after 12.43 mi. The final section is 1.36 mi. Onset is a section boundary, not an exact moment.

Full methodology & sources

Use the same complete-split and reviewed edition-quality cohort as the main analyses. Missing age or gender alone does not remove a finish. Rates describe recorded finishes, not unique people or physiological causes.

Marathons, years and recorded data. Chart samples may be smaller than the eligible analysis cohort.

Current study results, methods and source checksums (JSON)Published slowdown method (2021)Study methodology

Sustained slowdown by course

Percent (%)

Amsterdam26.2%

Auckland37.6%

Austin37.9%

Barcelona24%

Berlin24.4%

Boston21.5%

Chicago32.7%

Copenhagen27.5%

Dubai42.8%

Eindhoven24.3%

Gold Coast42.4%

Hamburg22.7%

Helsinki29.5%

London30.6%

Madrid38.8%

Melbourne35.1%

New York35.1%

Oslo32.6%

Paris33.9%

Prague42%

Rotterdam31.5%

Singapore47.3%

Stockholm26.1%

Sydney43.1%

Valencia18.9%

Vienna21.4%

Warsaw26.8%

Washington45.9%

View exact values and sample sizes

Group	Sustained slowdown by course	Observations
Amsterdam	26.2%	129,673
Auckland	37.6%	5,536
Austin	37.9%	7,621
Barcelona	24%	50,206
Berlin	24.4%	719,320
Boston	21.5%	183,814
Chicago	32.7%	252,393
Copenhagen	27.5%	125,695
Dubai	42.8%	11,885
Eindhoven	24.3%	24,753
Gold Coast	42.4%	37,236
Hamburg	22.7%	24,417
Helsinki	29.5%	7,853
London	30.6%	284,671
Madrid	38.8%	103,749
Melbourne	35.1%	416
New York	35.1%	712,178
Oslo	32.6%	8,351
Paris	33.9%	99,970
Prague	42%	19,105
Rotterdam	31.5%	139,923
Singapore	47.3%	8,877
Stockholm	26.1%	33,825
Sydney	43.1%	11,345
Valencia	18.9%	85,809
Vienna	21.4%	2,546
Warsaw	26.8%	23,152
Washington	45.9%	213,840

The study

When do sustained slowdown episodes begin?

Detected episodes begin at different recorded sections. These boundaries locate the first qualifying section, not an exact onset.

1,023,450 finishes with sustained slowdown

Full methodology & sources

The denominator is finishes with a detected episode. Distances are not proof that distance predicts slowing better than elapsed time.

Marathons, years and recorded data. Chart samples may be smaller than the eligible analysis cohort.

Current study results, methods and source checksums (JSON)Study methodology

First qualifying section starts at

Percent (%)

12.4 mi8.6%

15.5 mi19.4%

18.6 mi37.1%

21.7 mi34.8%

View exact values and sample sizes

Distance (mi)	First qualifying section starts at	Observations
12.4 mi	8.6%	1,023,450
15.5 mi	19.4%	1,023,450
18.6 mi	37.1%	1,023,450
21.7 mi	34.8%	1,023,450

How much does the definition matter?

Changing the amount or duration of slowing changes how many finishes qualify.

Comparison

Slowdown threshold and frequency

Percent (%)

Minimum distance

View exact values and sample sizes

Minimum slowdown (%)	Slowdown threshold and frequency	Observations
10	67.7%	3,328,159
15	53.5%	3,328,159
20	41.1%	3,328,159
25	30.8%	3,328,159
30	22.5%	3,328,159
40	11.7%	3,328,159
50	6.1%	3,328,159

Methodology & sources

At least 25% slower than the 3.11–12.43 mi reference pace for contiguous recorded sections totaling at least 3.11 mi after 12.43 mi. The final section is 1.36 mi. Onset is a section boundary, not an exact moment.

Each curve uses the same eligible finishes. These are descriptive thresholds, not physiological cutoffs.

Study methodology Current figure data, methods and source checksums

How does slowdown vary by recorded age?

Age groups have different observed rates. Age, fitness, conditions and field composition are not separated.

Comparison

Sustained slowdown by age

Percent (%)

Recorded gender

18–2929.1%

30–3926.4%

40–4924.7%

50–5925.9%

60–6926.2%

70–7924.5%

View exact values and sample sizes

Group	Sustained slowdown by age	Observations
18–29	29.1%	98,468
30–39	26.4%	149,392
40–49	24.7%	130,393
50–59	25.9%	64,787
60–69	26.2%	13,985
70–79	24.5%	1,274

Methodology & sources

Use exact recorded ages from 18 through 89, in the displayed age bands. Missing ages remain missing. Every cell requires 100 finishes.

Study methodology Current figure data, methods and source checksums

How does slowing compare with recorded performance?

Earlier performance and the year of an observed best provide two different comparisons.

Comparison

Slowing by earlier recorded best

Percent (%)

Under 3 hours11.2%

3–3:3017.9%

3:30–4 hours24%

4 hours or longer38.3%

View exact values and sample sizes

Group	Slowing by earlier recorded best	Observations
Under 3 hours	11.2%	34,176
3–3:30	17.9%	84,229
3:30–4 hours	24%	130,151
4 hours or longer	38.3%	275,767

Methodology & sources

The earlier-performance chart uses only the fastest eligible finish in the two strictly earlier calendar years.

The second chart is retrospective: select each candidate identity’s fastest eligible recorded finish, resolving ties by earliest year then record ID. Year zero includes that best. Unequal follow-up and selecting a best can explain the pattern; it is not a training effect or lifetime best.

Source identity groups pass ambiguity, gender, birth-year and duplicate-edition checks, but are not independently verified people.

Study methodology Current figure data, methods and source checksums