Even though it’s illegal to discriminate, algorithms are just using your ZIP code to "guess" your race and income anyway.
Because geographic boundaries often mirror historical segregation and economic divides, ZIP codes act as 'proxy variables' that absorb demographic data. This allows automated systems to effectively 'see' forbidden categories like race and make biased decisions without ever actually having that information in the model.
The Geography of Inference: ZIP Codes as Geographic Proxy Variables in Algorithmic and Automated Decision Systems
SSRN · 6373738
Automated decision systems increasingly influence consequential outcomes in areas such as credit underwriting, insurance pricing, and healthcare. In these systems, the choice of input variables can materially influence the relationships a model is able to learn and the outcomes it produces. Geographic identifiers such as ZIP codes frequently appear in applied analytics because many economic, environmental, and infrastructure conditions vary across locations. Although these identifiers were origi