Analysis of racket-sport database rankings for behavioural effects and other anomalies

Abstract

This dissertation evaluated whether behavioural factors produced systematic deviations between observed squash match outcomes and those predicted by the SquashLevels rating model, focusing on two candidate mechanisms: same-day fatigue and effort modulation in mismatches. An anonymised SquashLevels dataset (sq_combined.csv), precompiled via SQL joins across multiple relational tables, was analysed using deviationfrom-expectation metrics derived from pre-match player levels and observed match dominance. For fatigue, players with at least two competitively weighted matches per day (weighting ≥ 0.75) were examined using a paired first-versus-last design (n = 237,114 player-day pairs), and a Wilcoxon signed-rank test assessed whether later matches showed systematic underperformance; the change was statistically significant (p = 1.18 × 10−37) but small and positive (median ∆Deviation ≈ +0.015), providing no evidence of fatigue related underperformance under this proxy. For effort modulation, matches were analysed from the stronger player’s perspective and a log–log regression baseline related log(RATIO) to log(Level Ratio) (n = 3,826,051, R2 ≈ 0.294); residuals were compared across SquashLevels-aligned mismatch bands (1.0–1.5, 1.5–3.0, > 3.0), showing a negative median residual in extreme mismatches (approximately −0.050), consistent with reduced dominance relative to expectation. Overall, the findings partially supported the unifying thesis: fatigue effects were not practically evident within competitively weighted same-day sequences, whereas extreme mismatches exhibited systematic underperformance consistent with effort modulation, supporting the rationale for dampened rating sensitivity in highly unbalanced matches and indicating that behavioural adjustments were most defensible when targeted to contexts where the deviation signal was strongest.

Read the full thesis here.