Representative Sampling off the Rural Road: Lessons from a Sampling Design and Survey Implementation Effort in Rwanda

Stuart Sweeney, University Of California, Santa Barbara
Jessica Marter-Kenyon, University of Georgia
Sophia Arabadjis, City University of New York

Demographic surveys implemented in rural Africa often proceed without access to a reliable sampling frame. In the absence of such a frame, a rule of thumb is to select every xth house along a road. This may result in non-representative sampling and biased inference for several reasons. This paper reports on a novel survey sampling approach implemented in Rwanda and the associated construction of sampling weights. The intent of the survey is evaluate the impacts of Rwanda’s villagization policy on aspects of family formation and time to first birth. We develop a sampling approach that implements a spatial inhibition process combined with aggregate data on certain areal characteristics of Rwandan districts and sectors. After reviewing the sampling algorithm, we describe the resampling approach used to estimate own and cross probabilities of inclusion, which are then used in a Horvitz-Thompson estimator. We evaluate the bias and efficiency of our approach using simulated data and relative to other large-scale survey information (census and Demographic and Health Survey). The paper closes with a description of lessons learned during the implementation phase of the survey.

Keywords: Data and Methods, Population projections, forecasts, and estimations, Spatial Demography

See extended abstract.