Usage and acceptability of data normalization in baseball pitching.

Abstract

In baseball pitching biomechanics, kinetic values are commonly ratio 'normalised' by dividing by mass or mass*height to allow for comparison between athletes of different sizes. However, creating a normalised ratio variable should meet certain statistical assumptions. Our purpose was to determine if elbow valgus torque predicted by pitching velocity is influenced by normalisation using regression model comparison with and without normalised torque values. Motion capture data for youth to professional pitchers (n = 1988) were retrospectively analysed. Normalisation assumptions were tested by comparing linear regression models to analogous models with an intercept fixed at zero and by examining remaining correlations between the confounding variable and new, normalised variable. Both mass (p < 0.001) and mass*height (p < 0.001) normalisation did not remove their respective relationship with torque. After accounting for mass or mass and height, velocity predicted 10% of variance in elbow valgus torque, whereas velocity predicted 59% of mass normalised torque and 45% of mass*height normalised torque. Ratio normalisation does not fully account for anthropometric variables that differ across pitchers and leads to different conclusions in the magnitude of velocity's predictive effect on elbow valgus torque. Therefore, we recommend using regression model comparison to account for anthropometric variables in baseball pitching kinetic data.