Your training data can predict a race time
Medium confidence
With enough training logged, models can estimate your race times with errors from about 2 percent to about 8 percent, depending on the model and the runner. No single formula works best for everyone, and models built on lab-style fitness markers predict trail races less well than road races.
Why it works
Race times come down to three things the body does. VO2max is how much oxygen your body can use. Lactate threshold is the pace where easy effort tips into hard. Running economy is how much energy running costs at your pace. Your training data reflects all three. Pace, heart rate, volume, and intensity let a model estimate them and predict from there.
The evidence
-
The authors built a simple rule of thumb to estimate a runner's lactate threshold (the intensity at which blood lactate begins to rise exponentially) without lab blood tests. The rule sets the threshold at 60% of a runner's endurance running speed reserve. Tested on 50 recreational runners drawn from a list of 803 volunteers (two outliers removed), the heuristic estimated the threshold within an acceptable error range for about 87% of runners, just short of the authors' 95% acceptance target (an ideal estimator would reach about 98.8%). They conclude it serves the same purpose as the Dmax lab protocol while being far more accessible.
-
Using raw training and race data from more than 25,000 recreational marathon runners, the authors showed that critical speed (the fastest pace a runner can hold in a physiological steady state) can be estimated straight from everyday GPS data. The best model, built from 400 m, 800 m, and 5,000 m efforts, predicted marathon finish time with about 7.67% error. Runners raced the marathon at roughly 84.8% of critical speed on average, with faster runners holding about 93% and slower runners about 78.9%. Runners averaged about 87.6% of critical speed early in the race, and those who ran the first half above about 94% of critical speed (particularly faster than it) were more likely to slow by more than 25% in the second half.
-
The authors analyzed wearable-tracker data (Polar) from about 14,000 runners across 1.6 million sessions and roughly 20 million kilometers. They fitted a model with two personal numbers, an aerobic power index and an endurance index (how much sustainable power fades as a run gets longer), to each runner's best race performances in a season. For seasons with three or more races, the model's race times were within 2.0% of actual times on average, and the endurance index gives a practical stand-in for threshold-type markers such as the lactate threshold. Training volume and intensity in the 180 days before a marathon correlated with the indices, though direction of cause was not established.
n=14000
-
The review found 114 equations in 36 studies: 61 used training and anthropometric variables and 53 needed laboratory tests. Reported accuracy ranged widely (r2 0.10-0.99 for 68 equations; standard error of estimate 0.27-27.4 minutes for 19). Heterogeneity meant no single best equation could be identified, and course gradient, sex and weather were often left out.
-
Ten volunteers wore three accelerometers (on the foot, the shin, and the lower back) and ran on a treadmill at 13.5 km/h until they could not continue. A model built from all three sensors together predicted time to exhaustion best, explaining about 79% of the variation (R-squared 0.792) using 21 movement indicators. Any single sensor did worse: the foot explained about 63%, the lower back about 57%, and the shin about 56%.
-
In 130 recreational runners, the percentage of maximum heart rate reached at 14.5 km/h tracked running performance closely: it correlated with peak treadmill speed at r=-0.92 and with personal-best IAAF score at rho=-0.80 (a lower %HRmax at that speed meant a fitter runner). Blood-lactate markers did slightly worse (|r| 0.84 to 0.86 for treadmill speed, |rho| 0.70 to 0.77 for IAAF score), and other heart-rate parameters were weaker (0.24 to 0.70). %HRmax at 14.5 km/h was the strongest predictor in regression models, so the authors suggest a %HRmax-based test as a cheap, non-invasive alternative for predicting performance.
n=130
-
The authors built a minimal, universal model of running performance based on relative metabolic power, with four parameters fitted to a group of runners or to an individual: a crossover speed (close to the speed at maximal aerobic power), a crossover time separating short and long events, and endurance values for short and long events. The model described male and female world and national records and individual personal bests from 800 m to the marathon with mean errors of (often much) less than 1%. As an application, it derives personalized characteristic race speeds for different durations and distances.
-
Recreational runners raced the marathon below their estimated critical speed: about 88% of CS in the low-decoupling third, 86% in the moderate third and 82% in the high-decoupling third. Average decoupling was 1.16 and was first detected about 25 km into the race. The low-decoupling group finished faster (217 vs 239 minutes for the high-decoupling group) and decoupled later (33 vs 19 km). Adding decoupling to a model based on critical speed and the curvature constant cut marathon-time prediction error from 6.45% to 5.16%.
n=82303
-
The authors measured body composition, training, strength, and heart-rate recovery in 84 male amateur marathon runners (average age 41) from the 2013 Madrid Marathon. Their best prediction model combined body fat percentage, heart-rate change during recovery after the Ruffier squat test, and half-marathon time, and it correlated strongly with actual marathon time (r=0.77). Swapping in 10 km time instead of half-marathon time lowered the correlation to r=0.73.
n=84
-
Trail performance correlated most often with maximal aerobic capacity (r about -0.50), lactate threshold (r about -0.48), velocity at maximal aerobic capacity (r about -0.68) and running economy (r about -0.31). Single-variable regressions explained 48-60% of performance variation and multi-variable models 48-99%. The classic endurance model was a weaker predictor in trail running than in road running.
-
This classic modeling paper combined published values for three physiological factors (maximal oxygen uptake, lactate threshold, and running economy) to estimate how fast a marathon could theoretically be run. Many combinations produced predicted times faster than the 1991 world record of 2:06:50, and the fastest predicted time was 1:57:58 for a hypothetical runner with a VO2max of 84 ml/kg/min, a lactate threshold at 85% of VO2max, and exceptional running economy. Joyner concluded that either substantial improvements in marathon performance were physiologically possible or that current ideas about the limiting factors in endurance running needed refinement and testing.
Why we call confidence medium
Several modeling approaches predict race times from training data or physiological markers. Emig and Peltonen (2020) used wearable data from about 14,000 runners and predicted race times with about 2% error; Smyth and Muniz-Pumares (2020) estimated critical speed from the GPS training data of more than 25,000 recreational marathoners and predicted marathon time with about 7.7% error, and Smyth (2022) cut that to about 5.2% by adding late-race durability. A 2019 systematic review (Keogh) found 114 marathon equations with accuracy ranging from poor to near-perfect and no single best one, and a 2021 review (de Waal) found lab markers predict trail races less well than road races. Most studies cover specific groups or distances, which keeps the confidence here at medium. Joyner 1991 laid the physiological framework.
Where it applies
Healthy adults
Last reviewed Oct 14, 2026. See how we score.