Small area learning

Abstract

Small area learning uses semi-supervised methods to estimate quantities such as area means when labelled data are sparse. The framework accommodates different prediction algorithms and forms of covariate information, including area-level aggregates. The paper develops estimators and confidence intervals, and examines how residual correction affects bias and variance when working models are misspecified.

Publication
arXiv preprint
Katarzyna Reluga
Katarzyna Reluga
Assistant Professor