Gerrymandering Risk in Metrics of Income Inequality: Preliminary Results [conference paper]
Conference
ACM Conference on Fairness, Accountability, and Transparency (FAccT) - March 3-10, 2021
Authors
Jayant Gupta (Ph.D. student), Alexander Long (M.S. student), Shashi Shekhar (professor)
Abstract
Given a space partitioning, the goal is to characterize the sensitivity of inequality measures computed on aggregated data to the choice of spatial units. This problem is analogous to election gerrymandering. This problem is societally important since aggregated data are used in policymaking to protect individual and household confidentiality. This problem is hard because of the large number of possible space partitionings and the lack of mathematical characterization of inequality metrics in the context of aggregation over space partitionings. Previous work such as Modifiable Areal Unit Problem (MAUP) literature has noted the sensitivity of many statistics to the choice of space paritioning without providing bounds. In contrast, this work proposes bounds for two types of income inequality measures. The proposed bounds are validated via mathematical proofs and computational simulations using synthetic census like household level data.
Link to full paper
Gerrymandering Risk in Metrics of Income Inequality: Preliminary Results
Keywords
geographic information systems, data science