Tim Le

Powerlifting Performance Regression

Which factors best predict a lifter's best squat, bench and deadlift? GLS models on 1,154 IPF World University Cup records.

Role
Co-author
Timeline
Oct 2025 – Dec 2025
Team
2 people
  • R
  • nlme
  • tidyverse
  • GLS Regression

TL;DR

The problem

What is the most parsimonious combination of bodyweight, sex, age and home-country advantage that predicts a lifter's best squat, bench and deadlift?

How it works

  1. 1

    Collect

    Filtered the OpenPowerlifting dataset to World University Powerlifting Cup meets (2017–2019, 2021–2025).

  2. 2

    Explore

    Correlation analysis and diagnostics showed unequal variance between sexes.

  3. 3

    Model

    Univariate GLS models per lift with a sex-specific variance structure (nlme).

  4. 4

    Select

    Iteratively refined predictors, comparing BIC, RMSE and out-of-sample R².

Tech stack & why

Analysis

  • R + nlmeGLS with flexible variance structures.
  • tidyverse / broomClean data wrangling and tidy model outputs.
  • patchwork / corrplotPublication-quality figures.

Hard parts & key decisions

Heteroscedasticity by sex

Ordinary least squares assumes equal variance, but spread in lifts differs between men and women. GLS with a sex-specific variance function fixed the residual pattern.

Results

1,154
records
8
years of meets
3
lifts modeled

What I'd do next