Search

Variance estimation of surveys with complex sampling designs (stratified or multistage design) is biased if design effects are not taken into account. In this article we estimate the variance of some of the variables included in the 1986 health interview survey of Barcelona (survey of 2,205 households--clusters--that were stratified by municipal district and by family size). We present a comparison of variance estimates using the Jackknife method and using the standard method (the one used for simple random samples). Jackknife method incorporates sampling desing effects for estimating the variance, as homogeneous pseudo-clusters are formed. This method leads to more accurate results than the other method especially when calculating global population estimates. Differences between these methods are not very important, whereas cluster design effects are more important for certain variables.