Analysis of survey data

Abstract

Data from surveys should be weighted using expansion weights to avoid biased estimates for the inference population. Calculation of expansion weights is a stepwise procedure that can be thought of as allowing respondents to represent all eligibles from the population. Design will have an impact on the variance of the estimates. Furthermore, the impact may be to inflate or deflate the variance and is often variable-dependent. Accounting for the design in estimating the variance will require specialized analysis procedures and often specialized software.