|
Abstract:
|
The standard large-sample-based approach to deriving a two-sided confidence interval performs poorly for proportions that are close to zero or one. Alternative methods are suggested in the literature, but no one approach appears to be universally superior, and some of the approaches are difficult to implement in practice, especially when the data are collected via a complex sample design. Such problems are even more evident when one estimates certain summary measures for two-way tables, such as the Index of Disparity (Pearcy and Keppel, 2002), where the parameter estimated is defined to be non-negative, but estimates close to zero are common. In this paper we consider the practical implementation of some alternatives to deriving confidence intervals in these cases using statistical software for the analysis of complex survey data (WesVar).
|
