Why can two software packages disagree on a two-sided permutation p-value for an asymmetric null distribution?
StatisticsLevel: IntermediatePermutation experiment design
Why can two software packages disagree on a two-sided permutation p-value for an asymmetric null distribution?
Hints
- A two-sided tail rule is not unique.
- Check what statistic was actually permuted.
Answer
Two-sided extremeness has different conventions: twice the smaller tail or a chosen absolute-distance statistic. Inspect the definition and null distribution before comparing values.
Source and licence: SciPy — resampling and Monte Carlo inference · BSD-3-Clause · answer: Quantoros editorial