A marathon prediction caught our attention: it was almost exactly twice the half-marathon prediction.
Put those numbers side by side and they suggest almost the same pace for twice the distance. If your familiar training distances are much shorter than a marathon, that deserves a closer look.
The ratio alone does not prove a prediction is wrong. The app was not literally doubling the half-marathon time, either: the two estimates could draw on different runs and personal records. But it raised the right question: were we giving enough weight to the demands of the longer distance?
What a shorter run can tell us#
A common way to predict race times is Riegel’s formula. Its idea is straightforward: as the distance gets longer, the time increases by a little more than the distance alone would suggest. You are expected to slow down.
With the 1.06 exponent our shorter-distance predictions use, a hypothetical two-hour half marathon projects to roughly four hours and ten minutes for a marathon, rather than four hours.
That is useful as a starting point. The limitation is that the same rule does not distinguish between two runners with the same shorter-race result but very different training histories.
Imagine two runners with the same half-marathon time. One is regularly preparing for a marathon. The other has plenty of speed over a shorter distance, but much less experience sustaining an effort for several hours. Their half-marathon results alone do not settle that difference.
What the research adds#
We turned to Vickers and Vertosick’s 2016 study of recreational runners. The researchers collected race and training information from 2,303 runners and examined how well predictions matched actual performances.
They found that the familiar distance-based approach could be too optimistic for the marathon. Their alternative combined a previous race performance with typical weekly running distance.
The distinction between distances matters. In this study, the traditional approach worked reasonably well through the half marathon at a group level, but not for the marathon. That does not mean every runner’s half-marathon prediction was accurate.
In a comparison using 156 runners whose results were set aside when building the new models, the paper’s distance-based marathon prediction was at least ten minutes too fast for roughly half the group. With a previous race result plus weekly mileage, that proportion fell to about a quarter. Individual errors could still be substantial: an improvement across a group is not a guarantee for any one runner.
The two inputs answer different questions. A recent performance gives a reference for speed; weekly distance adds context about the amount of running behind it. Neither tells the whole story, but together they provide more information than the shorter result alone.
This is the idea behind our marathon update. Two runners with the same recent performance no longer have to receive the same marathon estimate when their recorded weekly mileage differs.
It is not a rule that more mileage guarantees a particular finish time. Nor does it measure long-run readiness, fueling or recovery directly. The study offers a population-based starting point, not a personal promise.
How the feature works now#
Apex Run considers the last six weeks of recorded running. For the marathon, it combines a recent whole-run performance with average recorded weekly distance. Full start-to-finish time is used, including pauses.
There is no requirement to mark a run as a race. An eligible marked race takes priority when available; otherwise, the app compares eligible ordinary runs and selects the strongest marathon projection. A newer easy run does not replace a stronger performance just because it happened more recently. A selected marathon-length run is used directly.
The half-marathon estimate keeps its familiar distance-based calculation, with more relevant inputs. Its existing 10 km minimum now also applies to personal records: a quick 1 km or 5 km PB cannot pull it faster. Supporting runs and records must come from the six weeks leading up to the estimate. A longer easy run does not take priority simply because it is longer, and we have not added an untested mileage penalty.
The 5K and 10K predictions keep their existing behavior. Each distance has its own evidence dots, showing the supporting performances rather than promising a level of accuracy.
The Race Predictor uses the same marathon calculation with information you enter: a recent run, its full elapsed time and your average weekly distance over the past six weeks. There is no runner type or formula setting to choose.
Still a guide, not a promise#
The research used race results. Apex Run also accepts ordinary training runs, and that adaptation has not been independently validated. If your recent running is mostly easy, it may not reveal what you could do in a race.
Complete records matter, too. Missing runs can make your recorded weekly distance look lower than it really is. Keep existing runs synced, check their distances and times, and include a recent race result if you have one. There is no need to train harder just to change a prediction.
Weather, hills, pacing and fueling still matter on race day. The goal is not to make every marathon estimate slower. It is to give the longer distance more context, and give runners a more useful starting point.
If you use the app, the prediction guide explains what the numbers mean.