Abstract:
|
Geosampling is a probability-based, multistage, sample design in which geographic areas are partitioned into nested layers of GIS grid cells that are used as sampling units and boundaries for fieldwork. Prior applications of this methodology have relied on a simplified sampling and weighting approach, in which sampling units defined by the second layer of GIS grid cells ("SGUs") were sequentially inspected for households and accepted into the sample only if deemed residential. This process ceased once the desired sample size was achieved. As the total number of nonresidential units was unknown, weights assumed a simple random sample and a uniform distribution of the population across grid cells. This paper reviews the sensitivity of weights and estimates to the simplified approach using data from a new geosampling-based survey in two states of Nigeria. In the new study, complete residential inspection of SGUs was possible for a random set of clusters. Weights are calculated using both the simplified approach and new methods, and results are compared. Potential bias and variance impacts of the simplified approach are discussed and recommendations are provided for future studies.
|