I don't think Lindley's paradox supports p-circling
3 days ago
- #p-values
- #statistics
- #hypothesis-testing
- The article discusses the arbitrary nature of the p-value threshold (p < 0.05) and critiques the practice of 'p-value circling', where researchers express skepticism about p-values just below this threshold.
- It explains that under the null hypothesis, p-values are uniformly distributed, meaning all p-values are equally likely, which challenges the notion that p-values just below 0.05 are inherently suspicious.
- The author explores the idea that p-hacking (e.g., adding more data to achieve significance) can distort p-value distributions, creating a bump near the threshold, but argues this doesn't strongly justify p-value circling.
- Lindley’s paradox is introduced as a potential justification for p-value circling, where high-power studies may find p-values near the threshold less likely under the alternative hypothesis than under the null.
- The author concludes that p-values should not be interpreted as direct evidence and suggests using other statistics like Bayes factors for assessing evidence strength, while acknowledging the complexity and ongoing debate around p-value interpretation.