Title: Simplified variance estimation for multistage sample surveys
Authors: Guillaume Chauvet - ENSAI-IRMAR (France) [presenting]
Abstract: Multistage sampling designs are commonly used for household surveys. If we wish to perform longitudinal estimations, individuals from the initial sample are followed over time. If we also wish to perform cross-sectional estimations at several times, additional samples are selected at further waves and mixed with the individuals originally selected. Even in the simplest case when estimations are produced at the first time with a single sample, variance estimation is challenging since the different sources of randomness need to be accounted for, along with the needed statistical treatments (correction of unit non-response at the household and at the individual level, correction of item non-response, calibration). We consider a bootstrap solution that accounts for the features of the sampling and estimation process. This bootstrap solution is usually conservative for the true variance, in the sense that the sampling variance tends to be overestimated. The proposed bootstrap is illustrated with examples.