Evaluating Measurement Error in Segregation: Algorithm-Assisted Bias Correction

• In Event: Quantitative Study of Race & Gender

Abstract

Residential segregation in the United States is widespread, has persisted over time, and threatens fair economic opportunity and social cohesion. Most commonly used measures of segregation rely on aggregate data that impose arbitrary definitions of local geography. Previous work demonstrates that segregation measures are sensitive to the particular aggregation. Using recent advances in redistricting software, we evaluate the bias and uncertainty created by these measurement choices. We sample alternative Census tract maps that follow Census guidelines, ensuring that maps are contiguous and meet certain population bounds. We then calculate segregation metrics for each map to construct probabilistic distributions of segregation indices. With these data we provide bias-corrected estimates of racial and economic segregation across U.S. cities and quantify the uncertainty induced by aggregation measurement error. We demonstrate the potential of these data by re-examining contemporary and over-time segregation, and re-investigate leading studies of segregation's impact using our estimates.

Authors

• Jacob Brown, Harvard University

• Christopher Kenny, Harvard University

• Tyler Simko, Harvard University