|
Activity Number:
|
77
- Contributed Poster Presentations: Biopharmaceutical Section
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 3, 2020 : 10:00 AM to 2:00 PM
|
|
Sponsor:
|
Biopharmaceutical Section
|
|
Abstract #313440
|
|
|
Title:
|
One- and Two-Sided Hypothesis Tests for a Single Proportion with Clustered Binary Data: Derivation, Type I Error, and Power
|
|
Author(s):
|
Meghan I Short* and Howard Cabral and Janice M Weinberg and Michael P LaValley and Gina Peloso and Joseph M Massaro
|
|
Companies:
|
UT Health San Antonio and Boston University and Boston University and Boston University and Boston University and Boston University
|
|
Keywords:
|
hypothesis testing;
clustered binary data;
beta-binomial distribution;
type I error;
one-sided
|
|
Abstract:
|
While several confidence interval methods for a single proportion with clustered binary data have been introduced, hypothesis testing methods have yet to be explicitly developed and validated. This study derives three hypothesis test statistics from existing confidence intervals with good coverage and length properties, and measures their one- and two-sided control of Type I error and statistical power in a range of scenarios using Monte Carlo simulations. All three hypothesis tests performed well for values of the parameter near the midpoint of the parameter space (0.5), and could have inflated Type I error rates when the value of the parameter was closer to 0 or 1, especially when sample size was small. The newly proposed Wilson Score Continuity-Corrected Test for Clustered Binary Data had the best Type I error control for both low- and high-sided one-sided tests in a range of scenarios, where the Edgeworth expansion and uncorrected Wilson Score intervals had greater asymmetry with regard to Type I error control. These results highlight the importance of validating performance of one-sided as well as two-sided hypothesis tests for data with bounded parameter spaces.
|
Authors who are presenting talks have a * after their name.
