Algorithmic Bias and Historical Injustice: Race and Digital Profiling
Abigail Matthew, Amalia Rebecca Miller, Catherine E. Tucker · National Bureau of Economic Research · 2024
This paper studies the implications of attempts at "ethnic-affinity" profiling on Facebook that reflects users' engagement with content on Facebook.Profiling by ethnic-affinity is highly correlated with Census estimates of population race by geography.However, more users were profiled as African-American in former slave states relative to the baseline population.This occurs because the targeting algorithm was better at identifying Black users through differentiated engagement with cultural content in these states.This implies that policies restricting the collection of racial identity data will be unsuccessful due to the existence of proxies, and that relying on proxies may introduce troubling biases.