Disparate Impact Ratio (EEOC four-fifths / 80% rule)

For each group, the disparate-impact ratio is its favourable- outcome rate divided by the privileged group's rate. A value below 0.8 is the standard legal indicator of adverse impact.

For each group, the disparate-impact ratio is its favourable-outcome rate divided by the privileged group's rate. A value below 0.8 is the standard legal indicator of adverse impact.

Usage

morie_fairness_disparate_impact(
  y_pred,
  group,
  privileged = NULL,
  favorable = 1
)

morie_fairness_disparate_impact(
  y_pred,
  group,
  privileged = NULL,
  favorable = 1
)

Arguments

y_pred

Vector of decisions/assignments, one per individual.

group

Vector of protected-attribute values (e.g. race).

privileged

The reference group. If NULL (default) the highest-rate group is used and a warning is attached.

favorable

The value of y_pred counted as the favourable outcome (default 1).

Value

A morie_fairness_result; headline value is the worst (smallest) ratio across groups.

A named list: value (worst ratio), ratios, rates, privileged, adverse_impact, threshold, warnings, interpretation.

Examples

pred <- c(1, 1, 1, 1, 1, 1, 1, 1, 0, 0)
race <- c("A","A","A","A","A","B","B","B","B","B")
morie_fairness_disparate_impact(pred, race, privileged = "A")$value
#> [1] 0.6
pred <- c(1, 1, 1, 1, 1, 1, 1, 1, 0, 0)
race <- c(rep("A", 5), rep("B", 5))
res <- morie_fairness_disparate_impact(pred, race, privileged = "A")
res$value # 0.6  (group B rate 0.6 / group A rate 1.0)
#> [1] 0.6
res$adverse_impact # TRUE
#> [1] TRUE