582 – Stratification and Other Survey Sampling Theory
Significance Testing for Two Cluster Samples with Identical Clusters and Different Units
R. Lee Harding
ICF International
Ronaldo Iachan
ICF Macro
Pedro J. Saavedra
ICF International
In comparing groups using data from a cluster sample, the analyst needs to take into account the design effect in order to calculate statistical significance. There are several ways of handling statistically a situation where the same clusters are drawn but different units are selected for two surveys. The simplest seems to be to use a jackknife, creating one set of replicate weights. Then one can produce an estimate of the difference between means or proportions, and use the jackknife to obtain the variance of that estimate. Unlike separate jackknife estimates, this approach can use information about the common sample clusters. This research used simulated data to compare several methods of testing for significant differences in proportions. The methods included jackknifes or Taylor Series that take into account the common clusters in the two samples.
