The computation of confidence intervals for survey estimates is a key step in assessing data quality. Yet, it is usually not tractable to find variance estimators which reflect complex sampling design features such as unequal selection probabilities, unit or item non-response. In particular, the case of rotating samples is not easy to deal with because of covariance terms across the sub-samples.

We present an approach for variance estimation in case of rotating samples which is both theoretically justified and easy to implement using standard statistical software such as SAS, SPSS, STATA or R. The proposed approach is able to treat both cross-sectional and longitudinal indicators as well indicators of net changes. It also takes into account weighting adjustments for unit non-response and calibration to external data sources.

At the end, numerical results based on EU-SILC (European Statistics on Income and Living Conditions) for Luxembourg are shown.