Abstract

Mapping socioeconomic indicators at fine spatial resolutions is increasingly important for informing public policy. This task is particularly challenging in countries of the Global South, where household survey data are available but up-to-date census data are often lacking, making remotely sensed covariates an important alternative source of auxiliary information. A substantial statistical literature has developed small area estimation and model-based geostatistical approaches that combine survey data with remote sensing data. However, their performance may depend on the accuracy of household geographical locations, covariates, and population counts. We present a review and comparative assessment of these methods through a simulation study using Maharashtra region, India, as a starting point. The simulation scenarios reproduce key features of the information available in different countries by perturbing survey locations and introducing uncertainty into remote sensing covariates and population counts. We assess how these perturbations affect estimation accuracy across a range of modelling approaches, from classical area-level models to unit-context and geostatistical models. The study aims to clarify the robustness and limitations of alternative approaches and provide practical guidance for mapping socioeconomic indicators when supporting information is imperfect.