How Accurate Are Marathon Predictions?
A race calculator gives you 3:30. Your watch suggests 3:24. Your training partner thinks you should aim for 3:20.
Which number should you trust?
With the Melbourne Marathon & other Autumn world majors approaching, many runners are trying to work out what pace to start at. A recent race, a watch prediction or a laboratory test can help. But each tells us something different, and none captures everything that determines what happens over 42.2 kilometres.
When I’m helping a runner choose a target, I want to know both what their fitness suggests and whether their preparation supports it.
For those who'd prefer to listen to the podcast we recorded on this topic, head here.
Start with a recent race, then ask about the circumstances
A recent race is usually my starting point because it gives us a demonstrated performance.
But context matters. Was it raced hard? Was the course accurately measured? Was it hot or hilly? Were you rested, or carrying fatigue from a heavy training week?
All else being equal, I’d give a well-executed half marathon more weight than a 5 km when predicting marathon performance. It’s closer to the target distance and provides more information about sustained effort. But a recent, hard 10 km may be more useful than an old half marathon, or one you deliberately ran easily.
If you have several recent results, compare what they suggest. A 5 km that predicts a much faster marathon than your half marathon does raises a useful question. Does that reflect your endurance preparation, your strengths over different distances, or simply different conditions and fitness between races?
I wouldn’t automatically choose the most appealing number.
What does a race calculator actually calculate?
One common approach is the Riegel formula. It scales your race time to another distance while allowing for the slowing that typically occurs as distance increases.
Using the common exponent of 1.06, a 90-minute half marathon produces a marathon estimate of approximately 3:07:39.
The calculator hasn’t seen your training, though. Two runners with the same half-marathon result receive the same estimate, even if one has trained consistently for months and the other has had an interrupted build with limited long runs.
Vickers and Vertosick studied 2,303 recreational runners. Their evaluation found that Riegel’s marathon predictions were at least ten minutes too fast for approximately half the marathon runners. Models incorporating weekly mileage and previous race performances improved prediction. That doesn’t mean everyone should simply add ten minutes. It shows why the training behind the result matters. Vickers & Vertosick, 2016
What about Jack Daniels’ VDOT?
Daniels’ VDOT uses a different model, incorporating relationships between running speed, oxygen cost and the proportion of aerobic capacity sustainable over different durations.
A race result gives you a VDOT score, which corresponds to equivalent performances at other distances. You don’t need a laboratory VO₂max test to use it.
“Equivalent” is the useful word here. Your half marathon may correspond to a particular marathon time, but your preparation and race conditions affect whether you can deliver it. V.O2 makes this distinction in its own explanation of equivalent performances. V.O2
Does the newer Valencia model improve prediction?
Oficial-Casado and colleagues used paired Valencia half-marathon and marathon results from 2022 and 2023 to develop a model using half-marathon time and sex.
Their reported average absolute percentage errors were 5.67% for Valencia and 7.92% for VDOT, averaging across finishing-time categories. VDOT performed better below three hours, while Valencia performed better in several slower groups.
Actual marathon time | Valencia model’s mean absolute percentage error |
3:00–3:30 | 3.85% |
3:30–4:00 | 4.10% |
4:00–4:30 | 4.76% |
4:30–5:00 | 5.56% |
For scale, 4.1% of a 3:45 marathon is about nine minutes. That isn’t a guaranteed ±9-minute prediction range. Testing within the Valencia dataset also doesn’t establish the same accuracy elsewhere.
Try the Valencia calculation
Use your half-marathon time in minutes:
Predicted marathon time = (2.28 × half-marathon time) − 11.70 + sex adjustment
Add 5.49 minutes for men or 0 for women.
A 90-minute half marathon therefore predicts approximately 3:19 for a man or 3:14 for a woman. The adjustment reflects this dataset, not an individual assessment of endurance.
Treat the result as a reference point. It doesn’t directly assess training, fuelling, durability or race-day conditions. Oficial-Casado et al., 2026
What can physiological testing tell us?
Joyner’s classic model brings together three major determinants of endurance performance: VO₂max, the fraction of that capacity a runner can sustain, and running economy.
In practical terms: how much oxygen can you use, how much of that capacity can you sustain, and how much energy does your running speed require?
These help explain why two runners with the same VO₂max can have different performances. A single VO₂max number doesn’t describe the whole runner. Joyner, 1991
For marathon prediction, there’s another question: how well do those characteristics hold up as the run continues?
Durability: what happens to your fitness under fatigue?
Durability describes how long physiological characteristics remain stable during prolonged exercise and how much they deteriorate.
I explored this topic in my previous podcast conversation with Michele Zanini here. It’s particularly relevant here because marathon performance depends on more than what you can demonstrate when fresh.
In a small study of 18 London Marathon runners, Hunter and Muniz-Pumares found that smaller declines in lactate-threshold speed after a 90-minute run were associated with faster marathons. VO₂peak and threshold speed declined on average; running economy and fractional utilisation did not significantly change in that protocol. Hunter & Muniz-Pumares, 2025
That doesn’t establish a universal durability test. It does support looking beyond fresh fitness.
In training, I’d examine how the later stages of long runs compare with the earlier stages. Does effort rise sharply at a similar pace? Was the run appropriately fuelled? Does the same pattern appear across comparable sessions?
Heart rate alongside pace can prompt those questions, but drift alone isn’t a diagnosis. Weather, terrain, hydration and sensor quality affect interpretation.
What training supports the target?
A fast tempo session can be reassuring. I’m more interested in how it fits into the whole build.
Has training been consistent? Have longer runs been manageable? Is the runner recovering between sessions? Have they practised fuelling at the effort they intend to sustain?
Muniz-Pumares and colleagues examined training preceding more than 150,000 marathon performances. Faster runners generally completed more volume, with much of the additional training at lower intensity. Time in the higher-intensity zones was comparatively similar across performance groups. Muniz-Pumares et al., published online 2024
This was observational research. It identifies associations, rather than proving that copying faster runners’ training causes faster results.
The researchers also estimated three intensity zones from critical speed. Those zones aren’t interchangeable with your watch’s five zones or individually measured lactate thresholds. Their lowest zone covers a broader range of easy running than many runners associate with “Zone 1”.
The finding doesn’t make tempo work unimportant. It does challenge the idea that a collection of impressive hard sessions tells us everything about marathon readiness.
How much weight should you give your watch?
A watch prediction is another estimate to consider, but validation needs to match the device and distance.
For example, a study of the Huawei Watch GT Runner examined predictions for 5 km, 10 km and half-marathon performances. It found strong agreement, alongside meaningful individual variation. It did not validate marathon predictions or establish the accuracy of other watch brands. Dai et al., 2025
I’d look at whether the prediction agrees with recent races and training. If it suggests a major improvement that nothing else supports, I’d investigate before adopting it.
The course and weather belong in the prediction
Your fitness may be unchanged, but the conditions can change what pace is realistic.
Large-scale marathon research links warmer conditions with slower performances, with the relationship differing by performance level. There isn’t one temperature adjustment that fits every runner. El Helou et al., 2012
Hills also change the energy cost of running. The arrangement of climbs and descents matters, not just the total elevation gain. Minetti et al., 2002
A target based on a cool, flat race needs reconsideration if race day presents different demands.
Turning the estimate into a race plan
I’d bring the information together rather than let one calculator choose the pace:
What do your recent races suggest?
How consistent and specific has your preparation been?
How have you handled longer efforts and fuelling?
Do the course and expected conditions support the target?
Agreement between these gives me more confidence. Disagreement tells me where to look more closely.
For runners approaching a marathon, this review should help shape pacing and expectations. It shouldn’t become a reason to squeeze in missed training or run a last-minute session to prove you’re ready!
A prediction is useful when it helps you make a better decision. The aim is to choose a pace that your fitness, preparation and race conditions support, then allow yourself to finish well!
For the podcast we recorded on this topic, head here.
