Methodology

This predictor is based on log-level regression models trained on >2,800 professionally timed men's and women's 100m races with 10m split data, which is available here. A separate model is fit for each combination of sex, block distance, and fly distance.

The data was cleaned to remove Paralympic athletes and races where the athlete was obviously injured. Reaction time was removed from the data to isolate the effects of the block and fly times. For races that didn't have reaction time available, the average reaction time for the dataset was used.

Fly times were converted to velocities (m/s) prior to training the models as I suspected velocity would be more predictive. A log-level regression was chosen as the data didn't look linear, and because the log-level model produced a significantly better fit than a linear model. The trained models have R² values between 0.92 and 0.99.

Wind correction is based on Moniat, Fabius, and Emanuel (2018).