No. 662 - Accounting for sampling design in the SHIW

This paper analyses how sampling design affects variance estimates and inference using the data collected by the Survey on Household Income and Wealth (SHIW). The SHIW combines three basic features: stratification, clustering, and weighting to correct for unequal probabilities of selection among sampling units. A model to assess variance is presented and a Jackknife Repeated Replication method is employed to estimate variance. Empirical evidence shows that: 1) simple random sampling formula for variance underestimates by a factor of between 3 and 2 the estimates that take into account all the design features; 2) the bias of unweighted estimates may be fairly substantial; 3) all these factors can seriously mislead inference based on SHIW data.