Predicted-vs-realised rank audit by demographic group

Usage

morie_fairness_predpol_calibration_audit(areas, mean_risk, outcome_rate, group)

Arguments

areas: Area identifiers (per area, not per record).
mean_risk: Mean predicted-risk score per area.
outcome_rate: Realised-outcome rate per area.
group: Majority/dominant group per area.

Value

morie_fairness_result; $value is the largest-magnitude per-group mean rank gap (positive = over-policed).

Examples

set.seed(6)
areas <- paste0("a", 1:8)
mean_risk <- seq(0.1, 0.9, length.out = 8L)
outcome_rate <- mean_risk + rnorm(8L, 0, 0.05)
group <- rep(c("X", "Y"), each = 4L)
morie_fairness_predpol_calibration_audit(areas, mean_risk, outcome_rate, group)
#> Predictive-Policing Calibration Audit
#> =====================================
#>   Areas audited                   8
#>   Spearman rho (risk vs outcome)  1
#>   Worst group rank gap            0
#>   Worst-affected group            X
#> 
#> Overall the ranking is well calibrated (Spearman rho = 1.00): predicted risk broadly tracks realised outcomes. No group's areas are systematically mis-ranked; the rank gaps are small across groups.