Inquiry Response: Notes on Assessing Decision-Making Models for Bias
We build models in-house models to determine lending for people who buy our products. We rely on these algorithms for auto-approvals, and we’re concerned about model bias. How can we address this issue?
THE BIG IDEAS:
- Disparate impact
- Demographic data
Regulations for bias in modeling are coming. Be prepared by assessing how well you measure model accuracy. For lending, in particular, it’s important to understand whether accuracy differs for particular sub-populations.