Any Way You Slice It: Racial Segregation Statistics are Robust to Aggregation Bias

Residential segregation in the United States is widespread, persistent, and threatens economic opportunity and social cohesion. Most segregation metrics, however, rely on aggregate data with arbitrary spatial boundaries. The sensitivity of segregation measures to those boundaries is well theorized but rarely measured. Leveraging modern redistricting software, we simulate millions of alternative Census tract maps that satisfy Census guidelines, compute segregation for each, and recover the full probabilistic distribution of common indices. These simulations yield new estimates of racial segregation across U.S. cities and, crucially, measure the aggregation‑induced variability hidden in conventional estimates. Encouragingly, values calculated with official tract definitions closely track the mean of the simulated distribution, exhibiting only a slight upward bias, and segregation metrics shift modestly across maps—though variability grows in smaller cities. While our findings confirm the practical robustness of Census tracts, our data and software provide a general framework for diagnosing and correcting spatial‑aggregation error in other contexts.