Using gridded population and quadtree sampling units to support survey sample design in low-income settings.

Area of Science:

• Demography
• Survey Methodology
• Geographic Information Systems (GIS)

Background:

• Household surveys in low- and middle-income countries (LMICs) rely on census data for sampling frames, but this data is often outdated or incomplete.
• Outdated or incomplete census data can lead to coverage issues in survey sampling frames.
• Gridded population datasets are increasingly used as alternatives for creating survey sampling frames when census data is unavailable.

Purpose of the Study:

• To implement and evaluate a novel quadtree algorithm for generating household survey sampling frames using gridded population data.
• To address limitations of existing methods, such as uniform grid cell (UGC) sampling, which are resource-intensive and can result in heterogeneous sampling units.
• To improve the accuracy and efficiency of creating sampling frames in contexts with limited or outdated census information, using Somalia as a case study.

Main Methods:

• A quadtree decomposition algorithm was applied to gridded population estimates to create a population sampling frame.
• The quadtree approach successively subdivides an area into four equal quadrants until homogeneity is achieved within each quadrant.
• The performance of the quadtree method was compared against uniform grid cell (UGC) approaches (1x1 km and 3x3 km) using homogeneity metrics.

Main Results:

• The quadtree algorithm produced significantly more homogeneous sampling units compared to UGC methods.
• Calculations of standard deviation and coefficient of variation indicated superior homogeneity for the quadtree approach at national and regional scales.
• The study demonstrated outstanding performance of the quadtree approach in creating a reliable sampling frame for Somalia.

Conclusions:

• The quadtree algorithm reduces manual effort in subdividing areas and allows for accurate calculation of sampling weights.
• This method enhances the precision of survey estimates by producing more homogeneous population counts within sampling units.
• The approach offers significant labor, time, and cost savings, with potential for broader application in survey methodology.