Journal of Official Statistics, Vol.20, No.4, 2004. pp. 705–723
Confidence Intervals for Proportions Estimated from Complex Sample Designs
Alistair Gray, Stephen Haslett and Geoffrey Kuzmicich
Abstract:Korn and Graubard (1998) have suggested a method for producing confidence intervals for proportions estimated from a sample based on a complex sample design where the proportions are either very small or very large, or the sample size is small. Their method uses the exact binomial confidence intervals but with the sample size modified by dividing by the estimated design effect for the proportion in question. Statistics New Zealand wanted to use this method for the 1999 Gaming Survey (which has a different design to that for which the Korn and Graubard method was developed) and carried out a bootstrap analysis to see whether the coverage properties of such intervals were similar to those from the Korn and Graubard method. The article presents the results of this analysis.
Keywords:Complex sample design, proportions, exact binomial confidence interval, bootstrap confidence interval, exchangeable bootstrap
Copyright © Statistics Sweden 1996-2018. Open AccessISSN 0282-423XCreated and Maintained by OKS Group