Can you trust marathon time predictors?

Why predictions run fast

Most calculators take a shorter race and stretch it to the marathon. They leave out things that matter over 42 km, such as how much you actually train.

The Riegel formula, used by many online calculators, predicted marathon times that were too fast by at least 10 minutes for half of recreational runners. Weekly training volume added information that the formula missed (Vickers & Vertosick, 2016).

A goal that is too fast sets you up to start too fast, which is the main route to hitting the wall. In men, wall rates rose from 26% to 40% in the years just before a personal best, when they pushed hardest (Smyth, 2021).

How the common methods compare

VDOT is the race-equivalence system from Jack Daniels’ training books. In 8,261 runners in Valencia, VDOT predicted marathon time from a half marathon with a mean absolute error of 5.9%, and accuracy varied by runner level (Oficial-Casado and colleagues, 2026). As an example, 5.9% of a 4-hour marathon is about 14 minutes.

Watch predictions. A small, unpublished dissertation found that a Garmin predictor was about 7.6 minutes too optimistic for the half marathon (GSU dissertation, 2024). That is thin evidence, but it points the same way as the formulas. See also how accurate your running watch is.

Half marathon × 2.19. A rule of thumb from a journalistic analysis of the Vickers data multiplies your half-marathon time by about 2.19. The research report behind this site flags this figure as unverified. As an example, a 2:00 half marathon gives about 4:23.

Critical speed: a better anchor

Critical speed (CS) is the line between hard-but-sustainable running and severe running. You can estimate it from best efforts in your normal GPS training data over 6–12 weeks, such as 400 m, 800 m and 5000 m, or from three all-out efforts of 2–15 minutes each.

In more than 25,000 runners, CS from everyday training predicted marathon time to within 7.67% (Smyth & Muniz-Pumares, 2020). The same study showed how much of CS runners can hold for a marathon:

• Recreational runners sustained 79–93% of CS, depending on ability.
• The report’s rule table breaks this down by finish time: roughly 79–83% for 5–6-hour runners, 84–88% for 3:30–4:30 runners and 90–93% for runners at 2:45 or faster.
• Starting the first half above 94% of CS predicted large slowdowns.

An example: if your CS is 12 km/h (5:00 per km) and you are a 3:30–4:30 runner, 84–88% of CS is about 10.1–10.6 km/h. That is roughly 5:41–5:57 per km, or a finish of about 4:00–4:11.

Durability sharpens the prediction

Durability means how little your thresholds and running efficiency decline during a long run. Across 82,303 marathoners, the gap between heart rate and pace started to widen (“decoupling”) at about 25 km. Runners with little decoupling raced closer to their CS. Adding decoupling to CS cut the prediction error from 6.45% to 5.16% (Smyth and colleagues, 2022).

In another study, London finishers’ lactate-threshold speed fell from 12.8 to 12.1 km/h after a 90-minute run, and the size of that drop tracked marathon time (Hunter & Muniz-Pumares, 2025).

A practical check, graded C because the cut-offs are inferred: compare heart rate against pace in the second half of your long runs with the first half, and watch the trend across your build. Our suggestion: if heart rate drifts up a lot at the same pace, be more cautious with your goal.

Adjust for missed training

Breaks matter. In 292,323 runners, a 7–13-day training break in weeks 3–12 before the race was linked to a 4.25% slower finish, and breaks of 21–27 days to nearly 10% (Feely and colleagues, 2023). The report’s rule adds about 4% to the prediction for a 7–13-day break, 5% if it fell 3–7 weeks out, and up to 10% for a 3-week break.

As an example, 4% on a 4:00 prediction is about 9.6 minutes, giving roughly 4:10.

Build a range, not a number

The research report behind this site combines these methods into one approach, graded B:

1. Estimate from several methods: CS with the share that fits your ability, a recent half marathon adjusted for volume and longest run, and VDOT.
2. Use Riegel only as the fastest plausible time. Ignore the raw watch prediction.
3. Show a band of about ±8%, wider if you run under 40 km a week or it is your first marathon.
4. Check the goal pace against CS. Flag it as aggressive if it is above your ability-matched share of CS or above 93% of CS, and pick a safer target.

As an example, ±8% around 4:00 runs from about 3:41 to 4:19. Our suggestion: aim for the middle of the range and pace the first half conservatively. The marathon taper and fuelling plan also shape where in that range you land.

What we don’t know

• Most evidence is from race-result databases, not trials. They show how well methods predict on average, not for you.
• The Garmin finding comes from one small unpublished dissertation, and the ×2.19 rule from a journalistic analysis.
• The ±8% band and the decoupling cut-offs are practical choices, not tested limits.
• Hot or hilly courses are not built into these predictors. See running in the heat.

Sources

• Vickers & Vertosick (2016). How well race-time formulas predict recreational marathon times.
• Smyth (2021). How recreational marathoners hit the wall: an analysis of 4.1 million race results.
• Oficial-Casado et al. (2026). VDOT prediction of marathon time from the half marathon in 8,261 runners.
• GSU dissertation (2024). Accuracy of a Garmin race-time predictor.
• FiveThirtyEight. Journalistic analysis behind the half marathon × 2.19 rule.
• Smyth & Muniz-Pumares (2020). Critical speed from training data and marathon prediction in over 25,000 runners.
• Smyth et al. (2022). Heart-rate decoupling and marathon performance in 82,303 runners.
• Hunter & Muniz-Pumares (2025). Drop in lactate-threshold speed after prolonged running and marathon time.
• Feely et al. (2023). Training breaks before a marathon and finish time.