Practical Guide to Small Area Estimation (SAE) of Poverty using Remotely Sensed Variables - Example of Bangladesh
Preface
Welcome to your practical guide to statistical modelling of poverty at small administrative areas, i.e. Small Area Estimation! In this book, we will demonstrate how to leverage remotely sensed data in combination with household surveys to produce spatially disaggregated poverty statistics. We will both introduce theoretical concepts, as well as walk you step-by-step through the estimation procedure in R. You are free to either follow the guide from the start (Chapter 1), or, if you are familiar with theory and want to dive straight into the estimation – skip to the practical part. The guide has two sections:
Section A: Processing of geo-spatial covariates, such as accessibility, nightlights intensity, demographic maps, and topography – which will be used as inputs into poverty modelling in the next section.
- Practical application starts in Chapter 5.
Section B: Small Area Estimation of poverty, including covariate selection, model fitting, and necessary data transformations.
- Practical application starts in Chapter 8.
Whether you are a researcher, statistician, policymaker, or international development practitioner interested in knowing the spatial distribution of poverty at small areas and getting hands-on with the data – this guide is for you! Though this example is focused on estimating poverty at small areas (upazilas) in Bangladesh, we encourage users to apply this method to their own country context where appropriate.
Pre-requisites and system requirements
A foundational background in statistics or econometrics.
Intermediate knowledge of the R programming language. We will assume the user has already downloaded and set up R and RStudio.
Section A: 35GB of space available locally on your machine to download geo-spatial datasets.
Section A: at least 8 GB of RAM, but ideally 16 GB and higher.
Note on computations in Section A
Given Section A involves heavy computation using large geo-spatial datasets, we recognise that some users may not have the system specifications required for this to run smoothly – but fear not. We prepared a csv file with ready harmonised geo-spatial covariates. So, if you are unable to run computations in Section A, just download BGD.zonal.new.version.csv and proceed directly to Section B.
Before the practical part
Section B requires you to request access to the Demographic and Health Surveys (DHS) data through this link: The DHS Program - Data. Detailed instructions are in Chapter 8 of Section B. Please ensure that you begin the data request two days before attempting the practical exercise.
Acknowledgments
This guide was developed by the Data Innovation Team at Oxford Policy Management, in partnership with the University of Southampton, under the Data and Evidence to End Extreme Poverty (DEEP) research programme, funded by FCDO.