Cyclic Permutation Test: A Nonstandard Exact Test For Linear Models
summary
Proposes the Cyclic Permutation Test (CPT) that gives exact finite-sample Type I control for arbitrary fixed design and arbitrary exchangeable errors whenever 1/α is an integer and n/p ≥ 1/α − 1 (e.g. α = 0.05 requires n/p ≥ 19). Constructs 1/α linear statistics by solving a linear system so their joint law is invariant under a cyclic permutation group, applies a marginal rank test to those statistics, and boosts power by optimizing row ordering with a genetic algorithm (simulations show comparable power and an exact confidence interval by inversion).
AI-written summary · about
chapters (11)
- 0:00Introduction and background
- 3:09Hypothesis testing context
- 5:56Literature review findings
- 9:18Challenges of exact testing
- 11:02Cyclic permutation test
- 13:34Constructing the test
- 19:32Power and efficiency
- 23:12Algorithmic implementation
- 27:17Simulation performance
- 30:42Reflections and future work
- 34:02Closing discussion and Q&A
