Risk tools and predictive policing
Differences between groups with intervals, over time and place; calibration of an area ranking and a score’s spread across groups; individual fairness; bounds under label noise, Chouldechova’s identity and the resolution of a ranking; the Oaxaca-Blinder decomposition; simulated records with a known disparity.
| Function | What it computes | Python and command line |
|---|---|---|
fairdf() |
Differences between groups in a risk tool’s decisions | |
tmpaud() |
Disparity metrics over time and place | |
calaud() |
Calibration audit of a predictive-policing ranking | |
scrdsp() |
A risk score’s spread across groups | |
lipaud() |
Individual fairness: a Lipschitz audit | |
nsbnd() |
A true base rate behind a noisily recorded one | |
impfpr() |
The false positive rate the other rates imply | |
rnkres() |
Which pairs a noisy ranking orders | |
oaxbd() |
Oaxaca-Blinder decomposition of a gap between two groups | |
fairsim() |
Simulated records with a known disparity | |
nordet() |
Noisy-OR detection by patrols |