A simplified general method for cluster-sample surveys of health in developing countries

Abstract

General guidelines are presented for the use of cluster-sample surveys for health surveys in developing countries. The emphasis is on methods which can be used by practitioners with little statistical expertise and no background in sampling. A simple self-weighting design is used, based on that used by the World Health Organization's Expanded Programme on Immunization (EPI). Topics covered include sample design, methods of random selection of areas and households, sample-size calculation and the estimation of proportions, ratios and means with standard errors appropriate to the design. Extensions are discussed, including stratification and multiple stages of selection. Particular attention is paid to allowing for the structure of the survey in estimating sample size, using the design effect and the rate of homogeneity. Guidance is given on possible values for these parameters. A spreadsheet is included for the calculation of standard errors.

PIP: Emphasizing methods for practitioners with little expertise and no background in sampling, this paper presents a set of guidelines to follow in planning cluster-sample surveys of appropriate size in developing countries without undue bias. A self-weighting design based upon the World Health Organization's Expanded Program on Immunization is employed. The paper covers the topics of sample design, methods of random selection of areas and households, sample-size calculation, and estimating proportions, ratios, and means with standard errors appropriate to the survey design. Extensions, including stratification and multiple stages of selection, are also discussed. Giving guidance on possible values, the authors pay close attention to allow for survey structure in estimating sample size, using the design effect and the rate of homogeneity. A spreadsheet is finally included to aid in calculating standard errors.