Authors: Katerina Brandt*, University of North Carolina - Chapel Hill
Topics: Field Methods, Quantitative Methods, Development
Keywords: geographic sampling, open data, impact evaluation, Zambia
Session Type: Paper
Presentation File: No File Uploaded
Impact evaluation studies, public health interventions, and randomized control trials rely on traditional random sampling methods to address confounding. In places where updated census information is unavailable, it can be difficult to select a random sample of households or individuals for a study design. This paper presents an efficient random geographic sampling approach which works around limitations of unavailable census or administrative data by leveraging openly available building footprint data from OpenStreetMap. This sampling method was developed to select households for an impact evaluation of improved cookstoves in two neighborhoods of Lusaka, Zambia. At the start of the study, there was no available list of households from which our research team could randomly sample to ensure exchangeability in treatment and control groups. Additionally, rapid growth and building construction in the study neighborhoods make conditions for by snowball sampling and convenience sampling difficult. Our approach began by randomly selecting building footprints from OpenStreetMap then dividing these selected buildings into geographic sections. Our enumeration team visited each of the selected buildings and recorded whether it was residential and currently occupied. If a building was currently occupied, the enumerator recorded contact information for at least one member of every household in a building. From this list of contacts, households were randomly selected to be included in the baseline survey. Compilation of the household contact list was carried out over 2 days to included 1800 households. The advantages and limitations of this approach to sampling and implications for geographic sampling methodology are discussed.