|
Abstract:
|
For proportions, we know how to prescribe the effective sample size required if the required estimate, its expected values, and the precision requirements are provided. After the data are analyzed, you may be asked if some estimates should be suppressed because they lack precision. Both problems require reasonable criteria which, for binary data, vary based on the true population values. Even if direct estimates are used, the logistic transformation may be used to construct asymmetric confidence intervals on the proportion scale which are bounded between 0 and 1. This paper explores criteria related to standard errors, relative standard errors, and standard errors on the logit scale. It is assumed that the survey will involve multiple estimates with a range of values of the true proportion.
|
