Income Segregation Analysis in Limited-Data Contexts: A Methodology Based on Iterative Proportional Fitting

Since the 1950s, researchers in Urban Geography have created multiple instruments for measuring income segregation. However, the computation of such indexes requires the availability of income data and population distribution for small areal units. This approach is problematic for countries and cities where a government's decennial census does not collect or report income data for small-enough areal units to capture income variability within a neighborhood. To address this gap, we use Iterative Proportional Fitting (IPF) to combine neighborhood-level census data with an individual-level income survey data and then estimate small area discrete and continuous income distributions for each small area. We show that it is possible to compute segregation indices based solely on estimated probability distributions without the need to generate a full synthetic population or to obtain integer population counts. We test our empirical method with the case of Mexican cities, for which global and local indexes of segregation are computed with bootstrapped confidence intervals. The major contributions of this article are twofold. First, it uses a method for income-data generation to measure income segregation. Secondly, it demonstrates a linkage between the computation of segregation measures based on probability distributions and the feasibility of computing them directly from the same IPF estimated distributions of income.