A Comparison of Methods for Estimating Confidence Intervals for Proportions in Clustered Surveys

• Activity Number: 249
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Survey Research Methods Section
• Author(s): Natalie Exner*+ and Marcello Pagano
• Companies: Harvard University and Harvard University
• Keywords: clustering ; survey design ; confidence interval

Abstract

In public health surveillance, surveys designed to estimate the prevalence of an outcome are often complex, including design features such as stratification, clustering, and/or disproportionate weighting. As a result, variance and confidence interval estimation can be challenging. Standard methods for constructing confidence intervals for a proportion include variance approximation methods used with a Wald-type interval and replication-based methods. Alternative methods include adaptations to the Wilson confidence interval. We evaluate these methods in a variety of settings, including low prevalence and small sample size settings. We make recommendations for which methods to use when estimating confidence intervals for proportions in different types of complex surveys. We use our findings to inform the design and analysis of an HIV drug resistance surveillance system for low- and middle-income countries.