In order to obtain the variance of an estimator – i.e. uncertainty of the information obtained from a dataset about the target population – that is statistically correct, the complex sample design, weighting and imputations must be taken into account. Due to anonymization, however, the published dataset does not contain all the necessary information about the sample design (e.g. strata or cluster) or weighting (e.g. design or nonresponse weights). This makes it impossible to leave the task of correct variance estimation entirely to the user of the dataset.
Therefore, resampling weights have been constructed for the HFCS in addition to the final household weights. These weights are based on a bootstrapping procedure taking into account the sample design. 1,000 household subgroups were randomly sampled from the HFCS gross sample in accordance with predefined criteria, the creation of the final household weights was repeated, and finite population corrections were applied. As a result, the resampling weights added to the published user database contain all the information necessary for a correct variance estimation by bootstrapping. To this end, up to 1,000 resampling weights are available to the user of the dataset.
For more detailed information on the creation and use of resampling weights in the first wave of the HFCS in Austria, see chapter 8 of the Methodological Notes on the HFCS for Austria.