Calculate Somers' d for ordinal association with confidence intervals
Try It FreeSomers' d is a measure of ordinal association between two variables that takes a directional (asymmetric) approach. Unlike Goodman-Kruskal gamma, Somers' d accounts for tied pairs on the dependent variable, making it a more conservative measure. It ranges from −1 to +1, where 0 indicates no ordinal association.
CrossTabs.com calculates all three forms:
Based on concordant (C) and discordant (D) pairs:
Where Ty = number of pairs tied on Y but not X
Somers' d is closely related to Kendall's tau-b and gamma. Gamma ignores all ties, tau-b corrects for both row and column ties, and Somers' d corrects for ties on the dependent variable only. For prediction problems, Somers' d is often preferred because it penalizes ties on the outcome variable.
Use Somers' d when you have a clear dependent variable (e.g., predicting satisfaction from income level). Use tau-b when both variables are symmetric and neither is clearly dependent. Somers' d is more appropriate for predictive contexts.
The row variable is X and the column variable is Y. Use d(Y|X) when predicting Y from X (the most common case). CrossTabs.com reports all three forms so you can choose the appropriate one for your research question.
CrossTabs.com uses a jackknife variance estimator to compute confidence intervals for Somers' d, which provides robust standard errors even for moderate sample sizes.