A three-stage construct validity assessment of the Statcast breaking ball taxonomy using PCA, kernel density estimation, and Bayesian hierarchical binomial models (n = 782k pitches, 2020-2025). After conditioning on movement, sliders and sweepers are statistically indistinguishable on all five pitcher-controlled process outcomes, failing discriminant validity. Partial predictive validity is retained for the sweeper label on contact suppression outcomes.

Are sliders and sweepers actually different pitches? Using Statcast data from 2020-2025 and Bayesian hierarchical models, I show that after conditioning on movement, sliders and sweepers are statistically indistinguishable on all five pitcher-controlled outcomes. The sweeper is better understood as an extreme region of slider movement space, with a real but narrow contact suppression advantage.

An advanced breakdown of breaking ball typology.