Analysis of sample designs and sampling errors of the demographic and health surveys

This report documents features of sample designs used in these surveys and presents a detailed analysis of sampling errors for distinct variables. The first part of the report describes sample designs of 48 surveys conducted under the DHS program, which conform to a relatively standard model. All sample surveys are designed using scientific probability sampling. Most samples use two-stage stratified designs that incorporate selection of area units in a single stage, followed by listing and selection of households to yield approximately 30 female respondents per area unit, based on an average of 300 area units. The second part of the report analyzes sampling errors and design effects from these 48 surveys for the total sample, as well as for urban-rural domains, sub-national regions, and various demographic and socioeconomic subclasses. A total of 37 identically defined survey estimates are analyzed, covering fertility, family planning, and child health and mortality. Findings included the following. The pattern of variation of design effects is consistent across countries and variables, with individual values by country and variable well predicted by a multiplicative model separating the country and variable effects. At the country level, overall design effect, averaged over all variables and countries, is around 1.5. Within a country, regional design effects relate uniformly to total sample design effects across diverse variables. Urban-rural and regional differentials in design effects are small, because similar sample designs and cluster sizes were used across those domains within each country. Includes extensive statistical tables. (Author abstract, modified)