Training data can predict race performance

Running performance can be predicted from training data using models that incorporate heart rate response, pace history, training load, and physiological markers, though prediction accuracy varies by distance and training level.

In plain English

For well-trained runners with enough history logged, models can predict race times within about 2 to 5 percent. They are less accurate for newer runners and for very long ultras.

Why it works

Race performance is a function of VO2max, lactate threshold, and running economy. Training data (pace, heart rate, volume, intensity) provides proxy measures for these physiological determinants, enabling statistical prediction.

The evidence

Why we call confidence medium

Multiple modeling approaches (critical speed, TRIMP-based, machine learning) have shown predictive value, but most studies are limited to specific populations or distances. Joyner 1991 provides the physiological framework; recent work (Emig & Peltonen 2020) validates big-data approaches.

Where it applies

Healthy adults

Last reviewed 2026-05-26. See how we score.