IOL: Statistical and Machine Learning Approaches to Modeling Dependent Data for Official Statistics

This Introductory Overview Lecture presents statistical and machine learning approaches for modeling official statistics. At a high-level, this lecture will introduce both area-level models and unit-level models and discuss methods for incorporating various sources of dependence. At the area-level, the sources of dependence considered include spatial, temporal, multivariate, and network (i.e., socio-demographic). At the unit-level the same sources of dependence are considered in the context of an informative sample. Several topics will be introduced in this talk including, amortized learning, variational methods, and tree-based models. Special emphasis will be placed on computationally efficient approaches and principled uncertainty quantification. Case studies will be used throughout the lecture to illustrate the applicability of various methods.