The Observation Theory EncyclopediaFrom TSKAboutBy kindBy chapterBy Lean fileLedgerProvenance

disparate impact ratio concept

DefinitionThe rate of a favourable decision in one group divided by the rate in another. Equation 0.37. Also disparate impact, four-fifths, per group.
ExamplePositive rates of 0.48 for one group and 0.60 for another give a ratio of 0.8.
BookData Mining as Observation, draft 0.2, commit f3914f0; entry id disparate-impact-ratio, kind concept.
Statusno ledger row names this entry. Corrections: none recorded.
Defining equation

Book equation 0.37.

Assumptions and scope
  • The ratio of the favourable-decision rate in one group to the rate in another. It is one exactly when the rates agree, it inverts when the groups are swapped, and the four-fifths rule is a statement about the rate gap.
  • The book’s verdict is per group with a null, the worst group’s false-clear rate over the groups large enough to score, which clears a bar exactly when every scored group does and is never below the pooled rate.
Prior artnone recorded
Evidencelean/DataMiningAsObservation/DisparateImpact.lean
Reviewednot yet reviewed; generated 2026-09-10 from records at the commits on the provenance page.
00.51stratumscoregroup Agroup Bgroup Cbarmean passes, the minimum fails
The positive rate of one group over another's.

Equation

Book equation 0.37.

\[\mathrm{DI}=\frac{\Pr[\hat y=1\mid g=a]}{\Pr[\hat y=1\mid g=b]},\qquad \text{equalized odds}:\ \mathrm{TPR}_a=\mathrm{TPR}_b,\ \mathrm{FPR}_a=\mathrm{FPR}_b.\]

Book equation 14.5.

\[\mathrm{FC}_g=\Pr\big[y=\text{violation}\ \big|\ \hat y=\text{clear},\ g\big],\qquad \text{verdict}=\max_{g:\ n_g\ge n_{\min}}\mathrm{FC}_g,\qquad \text{ABSTAIN otherwise}.\]

Conditions

Conditions are curated in entries.toml rather than read from a record.

Ledger

none

First stated

Chapter 0 section 0.18 and chapter 14 section 14.6 of Data Mining as Observation, with the per-group verdict of equation 14.5.

Measurements

none

Failures and corrections

none

Invariance envelope

none declared

Machine checked

lean/DataMiningAsObservation/DisparateImpact.lean, theorems ratio_eq_one_iff, ratio_swap, four_fifths_iff, verdict_le_iff, weighted_le_verdict, at observation-data-mining f3914f0; what the check covers is stated in the book’s appendix C.

Used in

Data Mining as Observation primer S, chapters 0, 14.

Related

equalized odds; min-over-strata; false-clear rate; abstention.

See also

Sources-table rows that share a record with the entry without naming it: chapter 14 section 14.1.

Status

Generated 2026-09-10 by encyclopedia/generate.py; book at observation-data-mining f3914f0; the commit of every record is listed in the encyclopedia’s provenance.

← discretizationdistance concentration →