A1212
Title: Small area estimation with big data covariates
Authors: Camilla Salvatore - Utrecht University (Netherlands) [presenting]
Angelo Moretti - Utrecht University (Netherlands)
Robin Markwitz - Utrecht University (Netherlands)
Abstract: Small area estimation (SAE) when auxiliary information comes from non-traditional, error-prone sources such as big data or non-probability samples is considered. While SAE methods typically improve precision by combining survey estimates with model-based predictions, their performance can deteriorate when covariates are measured with error, such as when they are based on big data sources. In such cases, standard area-level models like Fay-Herriot may yield estimates that are less efficient than the direct estimator. Building on the measurement error framework of Ybarra-Lohr, a simulation study assesses different error scenarios and proposes a spatial SAE approach that also accounts for dependence across areas. The model jointly incorporates measurement error and spatial correlation, allowing for information borrowing across neighboring areas while properly adjusting for uncertainty in the covariates. The approach is demonstrated in practice, for example when integrating survey data with sources such as Google Trends.
