Small area estimates for Voting Rights Act Section 203(b) coverage determinations

Abstract

The Director of the US Census Bureau every five years, under the Voting Rights Act (VRA) Section 203(b), determines which states and political subdivisions must provide ballot assistance in languages other than English. These rule-based determinations use small-area estimates of the proportions of voting-age citizen members of racial/ethnic Language Minority Groups (LMGs) who are limited Englishproficient (LEP) and of LEP LMG persons who are also illiterate. This large-scale Small Area Estimation (SAE) effort by the US Census Bureau is based on American Community Survey five-year data. This paper focuses on the unique attributes of this SAE problem, including the predominance in each LMG of tiny domains along with relatively few large domains that account for the bulk of LMG citizens. The data and small area models are treated separately for distinct LMGs, and the spirit of the law requires that the domain estimates should be produced in a similar way across LMGs for each type of geography (Jurisdictions, which are mostly counties, plus two types of American Indian and Alaskan Native areas). The paper describes and assesses the SAE models developed under unique constraints, including novel methodological aspects and the use of hybrid frequentist and Bayesian computational techniques; situates these modelling choices in the broader literature on SAE; and discusses the strengths and weaknesses of the resulting models.