This project presents a novel bootstrap-permutation-based alternative to the traditional two-way ANOVA. It aims to provide a flexible and assumption-light method to test for main and interaction effects when data violate common assumptions such as normality and homoscedasticity.
Classical two-way ANOVA relies heavily on assumptions like normality, equal variances, and absence of outliers. This project introduces a robust non-parametric approach that handles:
- Heteroscedasticity
- Non-normal distributions
- Interaction effects
- Outliers
- Bootstrap Resampling: Used to estimate effect sizes under minimal assumptions.
- Permutation Tests: Applied to generate empirical p-values for each main and interaction effect.
- Diagnostics: Includes visual tools for checking normality, homoscedasticity, and assumption violations.
/codes/: All R functions, test scripts, and diagnostics/data/: Real-world dataset used/report/: Final report and presentation slides/vignette/: A walk-through of how to use the method on example data
- Download Permova package (Permova_0.1.0.tar.gz folder )
- Install 'Permova' library in R
- Run the 'stratified_perm_test_sequential' function for your data
- Explore the vignette in
/vignette/for step-by-step instructions
- R
- Bootstrap & Permutation techniques
- Custom plotting functions
This project is licensed under the MIT License. See the LICENSE file for more info.