Confidence Intervals for Subsampled Poisson Random Variables

• Activity Number: 526
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Owen Devine*+
• Companies: CDC
• Keywords: Confidence Intervals ; Poisson ; Double Sampling

Abstract

The proportion of subjects with a given condition who exhibit an additional characteristic is often of interest in epidemiologic investigations. For example, the proportion of children with a specified birth defect with a secondary condition can be meaningful in understanding the etiology and sequlea of some conditions. In practice, the double level of sampling inherent in estimating these subsample proportions is often ignored leading to potential under coverage when using standard confidence intervals. I present a modification of Pearson-type confidence intervals under a Poisson assumption to reflect expected sampling variability in both the numerator and the denominator of the estimated sub-sample proportions. The performance of the suggested approach is evaluated via simulation across a range of scenarios consistent with birth defect-related epidemiology.