About the Short Course: 1 Day Course

Surveys conducted by National Statistical Agencies generally produce reliable estimates of population parameters when the sample size in the population subgroups of interest, called domains or areas (e.g. cities), is large enough. Modern users of survey data are increasingly fond of information at extremely fine levels. Standard estimation methods usually cannot meet these needs without drastically increasing the sample size and thus collection costs. Small area estimation methods address this issue by introducing model  assumptions.

This short course is divided into four main parts. In the first part, we describe the problem and some key concepts. In the second part, the most important of the workshop, we describe the well-known Fay-Herriot area-level model and the associated Empirical Best Linear Unbiased Predictor (EBLUP). Several model validation diagnostics are presented and illustrated using real survey data. In the third part, we briefly describe the Hierarchical Bayesian version of the Fay-Herriot model. Finally, in the fourth part, we present the EBLUP under a unit-level model. The objective of the course is not to present in detail all the small area estimation models and methods found in the literature. Emphasis is placed on the commonly used Fay-Herriot model and the validation of its underlying assumptions.

In-Person Event. Location Of Short Courses: University of Ottawa

Who is this course for?

Statisticians working in National Statistical Offices or other organizations, as well as academics, who are interested in the small area estimation problem.

Level of instruction: Intermediate

Learning outcomes :

The topics to be covered are:

• Direct vs indirect estimators
• Benefits vs risks of using small area estimation methods
• Description of the Fay-Herriot model
• Description of the Empirical Best Linear Unbiased Predictor
• Diagnostics and graphs for the validation of the Fay-Herriot model
• Hierarchical Bayes version of the Fay-Herriot model
• The EBLUP under a unit-level model

Description of course materials:

• PowerPoint slides
• Proposed delivery structure, including elements of engagement
• The instructor will deliver the course with time allotted to questions throughout the course

Knowledge assumed (prerequisites)

Basic knowledge of survey sampling and statistics, ideally some knowledge of linear regression.

About the instructor: Mr. Jean-François Beaumont

Jean-François Beaumont is Senior Statistical Advisor at Statistics Canada, where he has been working for more than 25 years. His role is to manage the Research and Development Program on statistical and data science methods and to provide consulting services for Statistics Canada's social and economic statistics programs. He is also the Editor of Survey Methodology. During his career, he has conducted research projects on various topics such as small area estimation, data integration, robust estimation, bootstrap variance estimation, treatment of missing values and other estimation and inference problems.

Affiliation: Statistics Canada