Understanding the use of weights in the analysis of data from multistage surveys

Understanding the use of weights in the analysis of data from multistage surveys. Large national surveys are powerful tools with which to examine a variety of important rehabilitation-related issues and are currently the only feasible method to study disability trends over time. Because it is impractical to draw simple random samples from the entire United States, national surveys, such as the Medicare Current Beneficiary Survey (MCBS), select random samples of subgroups of a population. Thus, respondents may have unequal probabilities of being included in the survey, and weighting must be used in the analysis before the results may be generalized to the entire United States. Surveys such as the MCBS are rich sources of data for rehabilitation medicine, and it can be expected that more research will be conducted using these data sources. Statistical analysis of these data should account for the sampling scheme used in data collection. We review the principles involved in the design of multistage samples, the calculation of weights, and their use in the data analysis, focusing on their importance in the estimation of population values. Our objective is to help readers to understand and interpret results of research articles using this methodology. Examples using the MCBS data are provided to clarify the concepts presented in the article.