(55) UNSUPERVISED LEARNING MODELS REVEAL POOR TEST VALIDITY OF THE NFL 225-LB BENCH PRESS FOR INTRA-POSITION VARIANCE AND DRAFT OUTCOMES

Purpose: This study evaluated the effectiveness of the NFL 225-lb Bench Press test (BPT) in predicting intra-position variance among NFL Draft prospects, assessing its validity as a measure of upper-body strength and power in American football players. Methods: Performance records from the NFL Scouting Combine (2015-2022) were analyzed, including only players who completed all physical performance tests - height (Ht), weight (Wt), bench press test (BPT), 40-yd dash (40Y), shuttle drill (SHD), 3-cone drill (3CD), vertical jump (VJ), and standing long jump (SLJ) - to ensure data integrity (n=1137). Quarterbacks & specialists were excluded. Derived metrics included momentum (p40Y, pSHD, p3CD) and maximal anaerobic power (SLJP). Principal component analysis (PCA) using Oblimin rotation and Horn’s parallel analysis was applied for all positions to perform position group (PG) clusters via k-means clustering. PCA was then performed within each PG to assess the contribution of individual performance variables to intra-position variance. Binary logistic regression (BLR) and linear mixed models (LMM) were built for each position, using calculated principal component (PC) values to predict relative draft position (RDP) and draft status (DS). Results: K-means clustering identified three PGs: linemen (BIG), mid-size (MID), and skill (SK). PCA and parallel analysis yielded three PCs for BIG and SK, and two for MID (adjusted λ≥1). Table 1 shows the results of the intra-PG PCA. PCA explained 75% (BIG), 70% (MID) and 99% (SK) of intra-PG variance. Momentum-related variables (PC-1), accounted for the most variance (33-51%) across all three PGs. BPT was a meaningful contributor only in BIG (PC-3; Strength), explaining just 13% of variance while having no significant effect on RPD or DS (p >0.05). All other metrics contributed meaningfully to at least one PC in each PG. Conclusions: These results show that BPT is a poor indicator of intra-position variance and suggests that increasing BP repetitions has no meaningful impact on NFL Draft outcomes for any position. The weak predictive capabilities of this test support the claim that the BPT has poor test validity in the context of American football talent identification. We propose that alternative strength and power tests should be tested and implemented in talent identification frameworks to improve testing battery validity and predictive performance. Practical Applications: In the present configuration of the NFL Combine’s physical performance testing battery, players should prioritize training for other physical performance tests which were identified as significant contributors to intra-position variance, DS, and RDP, along with maximizing college game performance and skill development. We suggest that tests such as the isometric mid-thigh pull or barbell load-velocity profiling be studied and implemented in future scouting combines to improve test validity for sport-specificity. Acknowledgements: None