Cluster sampling is commonly used in survey designs. It is a practical method and often necessary when resources are limited. However, multi-stage cluster sampling is less efficient than simple random sampling (SRS). Thus calculations of sample sizes or standard errors based on the assumption of SRS need to be adjusted, to account for the loss of efficiency, or design effect. The magnitude of the design effect depends on the size of the clusters and on the internal homogeneity of the clusters (intra-class correlation).

This presentation addresses two issues. First, most researchers are familiar with SRS sample size determination, which involves assumptions about desired confidence level, acceptable error level and likely outcomes. However, the design effect may be less well known. This talk introduces the design effect concept.

Second, for new surveys, the design effect size is often not known. Instead, researchers need to depend on data from other surveys in similar areas. This presentation also reviews a variety of research, indicating what typical ranges of design effect researchers may encounter, for example, in heath and nutrition surveys, substance abuse and education. The typical ranges can be applied to estimating sample size for new surveys.