link: https://datatab.net/tutorial/kendalls-tau range: [-1, 1]
- τ = 1 indicates a perfectly increasing relationship between the ranked variables.
- (i.e. you got the rank perfect!)
- τ = -1 indicates a perfectly decreasing relationship.
- τ = 0 means there is no relationship
summary
- This is a ranking metric: how well does your model-predicted ranking, correspond to the real ranking
- it’s preferred over Spearman’s correlation Coefficient since it’s more robust
- τ = (number of concordant pairs - number of discordant pairs) / total number of pairs
-
- The pair is called concordant if the ranks for both X and Y increase together,
- The pair is considered discordant if the rank for X increases and the rank for Y decreases.
-
- The best way to understand it is to go to this link and walk through the example: