GridSample: an R package to generate household survey primary sampling units (PSUs) from gridded population data

Area of Science:

• Demography and Population Studies
• Survey Methodology
• Geographic Information Systems (GIS)

Background:

• Household survey data are crucial for policy decisions and resource allocation, often relying on census data.
• Census data can be outdated or inaccurate, compromising the reliability of survey sampling frames.
• Gridded population data presents a viable alternative sampling frame for household surveys.

Purpose of the Study:

• To introduce the R GridSample algorithm for selecting primary sampling units (PSUs) using gridded population data.
• To demonstrate the application of GridSample in replicating a national demographic and health survey.
• To explore the potential of gridded population sampling for improving survey accuracy and small area estimates.

Main Methods:

• Developed the R GridSample algorithm for two-step PSU selection: sampling 'seed' cells and growing PSUs.
• Utilized gridded population data (WorldPop 2010) and geographic boundaries for sampling.
• Implemented stratification and oversampling (urban/rural) within the GridSample framework.

Main Results:

• Replicated the 2010 Rwanda Demographic and Health Survey (DHS) using GridSample with comparable PSU numbers.
• GridSample produced 75 urban and 405 rural PSUs, with a median PSU population of 612.
• The study demonstrated the feasibility of using gridded population data for complex survey designs.

Conclusions:

• Gridded population sampling is a promising alternative to traditional census-based methods, especially when census data is unreliable.
• The GridSample R algorithm facilitates research into gridded population sampling methods.
• Further research is needed to fully assess the accuracy and feasibility of various gridded population sampling approaches.