The evaluation of the health systems of nations has been gaining increasing importance among health managers. Among the evaluation tools for the performance of health systems, nationwide health surveys have been more and more used to evaluate the health status of the population and satisfaction with healthcare from the user's point of view. Most national health surveys do not use simple random sampling, either due to budget restrictions or because time constraints associated with data collection. In general, a combination of several probabilistic sampling methods is used to select a representative sample of the population, which is called complex sampling design. Among the several sampling techniques, the most frequently used are simple random sampling, stratified sampling and cluster sampling. As a result of this process, the next concern is the statistical analysis of the data from complex samples. This paper deals with issues related to data analysis obtained from surveys using complex sampling designs. It discusses the problems that arise when the statistical analysis does not incorporate the sampling design. When the design is neglected, traditional statistical analysis, based on the assumption of simple random sampling, might produce improper results not only for the mean estimates but also for standard errors, thus compromising results, hypothesis testing, and survey conclusions. The World Health Survey (WHS) carried out in Brazil, in 2003, is used to exemplify complex sampling methods.
Survey; Sampling; Complex design; Statistical analysis; Brazil
