I am trying to understand the relative behavior of the following rank correlation statistics:
- Spearman coefficient
- Kendall Tau / Concordance percentage
- Normalized Gini coefficient (area under curve of percentage captured versus percentage observations)
- Normalized Area under ROC curve (for binary classifiers)
Are there good intuitions behind/discussions of relative (across datasets) or absolute (for a given dataset) differences for (any pair of) these measures?