Editing Cluster sampling (section)

==Disadvantages==

* Higher [[sampling error]], which can be expressed by the [[design effect]]: the ratio between the variance of an estimator made from the samples of the cluster study and the variance of an estimator obtained from a sample of subjects in an equally reliable, [[Simple random sample|randomly sampled]] unclustered study.<ref>Kerry and Bland (1998). [http://www.bmj.com/cgi/content/full/316/7142/1455 Statistics notes: The intracluster correlation coefficient in cluster randomization]. ''British Medical Journal'',  316, 1455–1460.</ref> The larger the [[intraclass correlation]] is between subjects within a cluster the worse the design effect becomes (i.e.  the larger it gets from 1. Indicating a larger expected increase in the variance of the estimator). In other words, the more there is heterogeneity between clusters and more homogeneity between subjects within a cluster, the less accurate our estimators become. This is because in such cases we are better on sampling as many clusters as we can and making do with a small sample of subjects from within each cluster (i.e. two-stage cluster sampling).
*Complexity. Cluster sampling is more sophisticated and requires more attention with how to plan and how to analyze (i.e.: to take into account the weights of subjects during the estimation of parameters, confidence intervals, etc.)