A composite binomial model derived from correlated random variables.
Groff-JH; Song-R; Schlecht-PC
Commun Stat Theory Methods 2000 Jan; 29(5-6):1179-1195
Motivated by a correlation problem in a power calculation of proficiency testing, a composite binomial model is developed to describe the distribution for the number of outliers from measurement results of multiple analytes contained in a single sampler. This model is different from other binomial models in that its component Bernoulli variables are derived from correlated random variables. By modeling the correlated random variables and selecting a binary link function to convert these correlated random variables to Bernoulli variables, various composite binomial distributions can be derived. Formulas for calculating the probabilities of the composite binomial distribution are provided. Although .the composite binomial model is developed for a power calculation, it can be applied to other problems related to a sum of correlated Bernoulli variables.
Mathematical-models; Models; Samplers; Sampling-methods; Analytical-models; Analytical-methods;
Author Keywords: Outlier; Bernoulli variable; Exchangeable; Link function
HGO/NIOSH, 4676 Columbia Parkway, Cincinnati, OH 45226, USA
Communications in Statistics - Theory and Methods