Small-Area Estimates as Covariates: A Measurement Error Approach

• Activity Number: 351
• Type: Topic Contributed
• Date/Time: Tuesday, August 11, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Government Statistics Section
• Author(s): Susanna M. Makela* and Andrew Gelman
• Companies: Columbia University and Columbia University
• Keywords: small area estimation ; sample surveys ; survey weighting

Abstract

Local- or neighborhood-level effects on individual health are often of great interest in epidemiologic and public health studies. In countries where population-based surveys are the main source of nationally representative health data, information from the individual or household level is often aggregated up to generate neighborhood-level variables. However, sample sizes at the neighborhood level are often small, and neighborhood-level characteristics are thus based on simple means or proportions from small samples of generally less than 30 observations. Obtaining more accurate estimates of these characteristics is particularly important when their effects are of substantive interest in the model. We propose a measurement error approach to simultaneously model the unobserved neighborhood-level characteristics and their relationship with the outcome of interest. We use simulation to demonstrate that this approach results in more accurate estimates of neighborhood-level effects than a simpler model using the direct mean. We illustrate with an application to a study of the effect of household- and neighborhood-level open defecation on child health and mortality in India.