Why can tiny floating-point differences change a signed-rank test on theoretically tied differences?
StatisticsLevel: IntermediateDistribution checking
Why can tiny floating-point differences change a signed-rank test on theoretically tied differences?
Hints
- Ranks depend on exact ties.
- Use measurement precision.
Answer
They can break ties and alter ranks. Compute differences carefully and apply domain-justified rounding before ranking, not arbitrary rounding chosen after seeing significance.
Source and licence: SciPy — distribution diagnostics and robust tests · BSD-3-Clause · answer: Quantoros editorial