The majority of nationwide face-to-face surveys are conducted in multistage cluster samples which means that respondents are clustered with small geographic areas (or sampling points). The conventional computation of standard errors and common test procedures are based on the assumption of independent and identically distributed observations, an assumption that is violated with clustered samples. One measure of the effect of the violation of this assumption is the so-called design effect, or the ratio of the variance of an estimator for a given sample design to the variance of an estimator for a simple random sample (Ref. 1). A large sample size is required to obtain the same precision as one would have had from a simple random sample. So steps are to be taken in advance to reduce design effects naturally arising by understanding the sources of such variations.
展开▼
