Half Marathon to Marathon Predictor: Which Number to Trust

Double your half marathon time and add somewhere between 10 and 20 minutes. That is the entire conversion, and it is what a half marathon to marathon predictor computes behind the interface — the disagreement between tools is only ever about the size of the addition. Feed a 1:35 half into four common exponent choices and the marathon estimates run from 3:18:04 to 3:30:49: a 12-minute, 45-second spread on identical input.

Which of those numbers you get depends on a decision the calculator made for you, silently, before you typed anything.

The ×2.085 rule is Riegel with the numbers already filled in

Riegel's formula is T₂ = T₁ × (D₂ ÷ D₁)^1.06. A half is 21.0975 km and a marathon is 42.195 km, so D₂ ÷ D₁ is exactly 2. The whole formula collapses to one multiplier: 2^1.06 = 2.0849.

That is where the ×2.085 figure circulating on forums and in blog posts comes from. It is not a competing model or a rule someone reverse-engineered from race results — it is Riegel evaluated at exactly double the distance, rounded to three decimals. Any tool that turns a half into a marathon is picking an exponent, whether or not it tells you which one.

Raise the exponent and the multiplier climbs quickly, because it sits in the power rather than the coefficient: 1.06 gives ×2.0849, 1.08 gives ×2.1140, 1.10 gives ×2.1435, and 1.15 gives ×2.2191.

One half time, four marathon predictions

The spread is not noise. It is always 6.44% of the fastest estimate in the row, because 2^0.09 = 1.0644 regardless of what time you feed in. Slower runners see a bigger gap in minutes for the same modelling uncertainty — 10:44 at a 1:20 half, 18:07 at a 2:15 half — which is why a five-minute argument about exponents matters more the further back in the field you start.

In pace terms the gap is smaller than it looks. For the 1:45 row, the difference between the 1.06 and the 1.10 prediction is 8.8 seconds per kilometre: 5:11.3 against 5:20.1. That is a real difference over 42 km, but it is close enough that a sloppy input will swamp it entirely.

Choosing your exponent before race day, not at km 32

Pick the column you can defend from your training log, not the one you like:

• 1.06 to 1.08 — the half was raced flat out within the last eight weeks, your longest run since then is 30 km or more, and you have held 60+ km per week for at least three consecutive weeks.
• 1.08 to 1.10 — longest run between 24 and 30 km, weekly volume between 40 and 60 km.
• 1.10 to 1.15 — first marathon, longest run under 24 km, or weekly volume under 40 km.

These bands are a planning convention, not a measured population relationship. Their job is to make you commit to a column in advance rather than discover one at km 32, and to give you a defensible conservative starting point when your training does not support the optimistic end. If you want the mechanics of how the underlying models differ, the longer breakdown of how marathon pace predictors work covers VDOT-based approaches alongside the power-law ones.

The input breaks the prediction more often than the exponent does

Four conditions have to hold before the half time is worth extrapolating at all. It has to be a race, not a tempo run with a bib on. It has to be recent — eight weeks is a reasonable ceiling, because you are predicting from fitness you may no longer have. The course has to be honest: net-downhill and point-to-point races with a tailwind produce times that flatter you, and the formula has no way to discount them.

The fourth condition is the one runners skip. Your second 10 km of that half should be within about 60 seconds of your first. If it was 90 seconds slower or worse, you did not race a half — you raced 15 km and survived 6, and doubling that time carries the pacing error forward into a distance where the survival section is 20 km long instead of 6. Pull the splits from that race before you trust the output; a half marathon splits calculator will show you what an even effort should have looked like next to what you actually ran.

What a finish-time predictor cannot see

A tool that takes one finish time has exactly one field. It has nowhere to put the 380 m of climbing between km 28 and km 34, nowhere to put a start line at 1,500 m of altitude, nowhere to put a forecast of 24 °C at 11:00, and nowhere to put the fact that a half is short enough for most runners to finish on the fuel they started with while a marathon usually is not. Whatever those things do to your day happens outside the number.

Measurement is the quiet one. Courses are certified at 42.195 km; watches routinely disagree. If yours reads 42.60 km at the finish, you ran 0.405 km more than the course, which at 5:11/km is 2 minutes and 6 seconds of extra running — larger than the gap between two adjacent exponents in the table above. Chasing the watch's instant pace instead of the cumulative clock at each marker is a more expensive mistake than picking the wrong column.

Why a half is a cheaper input than a 10K

Extrapolating from a half doubles the distance. From a 10K it multiplies by 4.2195. Because the exponent is a power, identical uncertainty costs far more from the shorter race: moving from 1.06 to 1.15 widens the marathon estimate by 2^0.09 = 6.4% when you start from a half, but by 4.2195^0.09 = 13.8% when you start from a 10K. For a runner around four hours, that is roughly 15 minutes of model uncertainty against roughly 33.

That is the practical case for racing a half in the build-up rather than relying on 10K sharpness — the same formula gets to be wrong by half as much. If a 10K is genuinely all you have, the 10K to marathon predictor walks through how much wider the resulting range has to be.

The seven-minute decision you make before km 21

Take the 1:45 row. The optimistic column (3:38:55) puts halfway at 1:49:28. The conservative one (3:53:01) puts it at 1:56:30. Seven minutes and two seconds separate them, and you spend that difference in the first half whether you decided to or not.

Come through 21.1 km at 1:56:30 and you have written off the 3:38 — you cannot get it back — but everything from about 3:50 down is still reachable on an even or slightly negative second half. Come through at 1:49:28 and only the 1.06 and 1.08 finishes stay compatible with even pacing; landing on 3:53:01 from there means running the second half 14 minutes slower than the first, which is what a blow-up looks like on a results page.

Pick your column, put its halfway split on the pace band next to the optimistic one, and let the second half decide which of them you had earned.