Use Pearson for a roughly linear numeric relationship; use the Spearman specialist for ordinal or monotonic relationships and outlier-resistant rank analysis.
Add rank correlation, best-fit line, RMSE and residual outlier evidence to the paired dataset.
| Metric / point | Value | Evidence |
|---|
Capture the current inputs, analysis settings and visible result as stable fingerprints, then flag structural method-fit issues that can be checked from the entered data. This complements the page’s existing mathematical audit; it does not prove distributional assumptions or causal validity.
Use descriptive context first, choose the method that matches the design, then keep inputs, settings and result evidence reproducible instead of copying a p-value without its analysis recipe.
Keep the existing Pearson/Spearman and regression audit, then test how stable the conclusion is. This companion computes a Fisher confidence interval for Pearson r, compares Pearson with Spearman, finds the observation whose removal changes r the most, and renders the paired data with the least-squares line.
| Point | X | Y | r without point | |Δr| |
|---|
The Fisher interval is an approximate interval for Pearson correlation and is not a substitute for study-design or assumption review. Correlation does not establish causation.
Use plots and descriptive summaries before formal tests. Check the sampling/design assumptions that matter for the chosen method, report effect size and uncertainty where available, and avoid treating a threshold such as p < .05 as proof of importance, causation, or truth. Browser calculations are educational/planning utilities, not domain-specific professional advice.
WebToolArc