Multistage Sampling

Objectives

When conducting a survey, it is often the case that a sampling frame, i.e. a list of population units, is not available. In this case, multi-stage sampling designs are often used, in which samples are selected in a nested fashion. Household surveys, for example, are often carried out by selecting communes, then blocks, then finally households.
In this type of survey, sampling is a complex stage, as several draws are necessary. This complicates both the calculation of estimators and the estimation of their variance. By its very nature, this selection procedure introduces dependencies into the observations, which also complicates non-response correction and survey data analysis.
The aim of this course is to present multi-stage sampling methods, and to understand its particularities. Simple variance estimation tools using resampling methods, for example, will be presented, and the specificities of nonresponse treatment for this type of survey will be discussed. Skills acquired :
Select and implement a sampling strategy for clustered data.
Produce estimators and associated precision indicators.
Possibly: Choose and implement a strategy for handling total nonresponse.

Course outline

Part 1: Introduction
Reminders on finite population sampling
Two-stage sampling: sampling and estimation
Cluster sampling: sampling and estimation

Part 2: Accuracy of two-stage sampling
Direct variance estimator
Simplified estimators
Bootstrap estimation

Part 3: Estimation for a two-stage survey
Treatment of total non-response
Calibration at household and individual level
(Analysis of survey data)

Prerequisites

Survey theory, linear regression, generalized linear model