|
Abstract:
|
While all complex surveys are designed to achieve a specified level of analytic precision, the NHANES survey has three important aspects that complicate analyses: the number of degrees of freedom is small, some outcome measures are based on subsamples of the original selection, and there are a number of important measured health characteristics that push the analytic needs for the survey beyond its original design specifications. While some analytic issues can be handled by combining multiple years of data, many of the statistical methods adopted for complex survey data are based on asymptotic results and are not adequate for rare outcomes. This presentation summarizes recent CDC research on exact confidence intervals for small proportions and percentiles. Also, test statistics---adjusted for small degrees of freedom---for comparing rare events are explored.
|
