Somers' D (asymmetric ordinal, Dxy + Dyx)

Coming soon

Advanced & specialized

This test is implemented and is currently going through StatMinds’ production verification: every statistic is independently checked against a trusted reference (scipy / R), locked with regression tests, and the screen is exercised across assumption-met/violated and significant/non-significant scenarios before it opens up.

See what’s live now

Asymmetric ordinal association coefficient.

D(Y|X) = (C − D) / (N₀ − Ty) — how well x predicts y's rank. D(X|Y) = (C − D) / (N₀ − Tx) — how well y predicts x's rank. They differ when the marginal tie totals on the two axes differ (Tx ≠ Ty). Reports both directional forms with asymptotic z + 95% CI, plus the symmetric harmonic form. The pipeline surfaces a primary-voice qualifier when |D(Y|X) − D(X|Y)| > 0.30 — substantial directional asymmetry that the symmetric form would obscure.

Worked example

How well does one ordinal variable predict another (directional)?

Somers' D is a directional ordinal association — the asymmetric counterpart to gamma, treating one variable as the outcome.

Result

The predictor accounted for ordinal variation in the outcome, Somers' D = .31, p < .001.

How you'd report it (APA)

Somers' D indicated a moderate directional ordinal association, D = .31, p < .001.

When to use it

  • Ordinal predictor → ordinal outcome (directional question)
    A clinician cross-tabulates pain-severity tier (mild / moderate / severe / very severe) against treatment-response category (no response / partial / full).
  • Ordinal logistic regression companion / AUC for binary outcomes
    A diagnostic-test developer computes D(Disease | TestScore) for an ordinal screening test.

When NOT to — use instead

Hypotheses

H₀: D_xy = 0 (and D_yx = 0) — no ordinal predictive association between x and y.
Hₐ: D_xy ≠ 0 (or D_yx ≠ 0) — a non-zero ordinal predictive association (two-sided default).

Parameter tested: population Somers' Dxy and Dyx

Assumptions (and what to do if they fail)

|Dxy| and |Dyx| differ when the marginal tie structure of x and y differs.medium

Check: Report Dxy, Dyx, and |Dxy − Dyx|.

If violated: Identical Dxy and Dyx indicate symmetric marginal-tie structure — in that case τ-b or γ may be equivalent and more interpretable.

Dxy, Dyx, γ and τ-b agree in sign.medium

Check: Report all four; sign disagreement flags non-monotonicity.

If violated: Sign disagreement between the four measures indicates a non-monotonic pattern.

Ready to run a Somers' D (asymmetric ordinal, Dxy + Dyx) on your own data?

Guided setup, automatic assumption checks, effect sizes, figures and an APA write-up.

Run this test →