Electronic Theses and Dissertations
Bayesian and Maximum Likelihood Estimation for Correlated Binary data Based on the Double Binomial Distribution
Date of Award
Doctor of Philosophy
In the statistical analysis of binary data, usually the binomial distribution is themost commonly used probability model in many applications. The binomial is a very usefuldistribution in many practical applications due to its simplicity and having a parameter thathas an intuitive meaning. However, it has a unique feature that is the variance depending onthe mean, an assumption that does not reflect reality in many practical applications especiallyin cases where the data exhibits greater variability than predicted by the distribution. Thetwo-parameter double binomial model introduced by Efron (1986) may be considered as auseful alternative to the one-parameter binomial distributions, given that it can account forboth overdispersion and underdispersion. In this dissertation, we obtain maximum likelihoodestimates for the double binomial distributions. The Bayesian methodology is also consideredfor estimation procedures and we demonstrate that, under the model assumptions, the fullyconditional posterior distributions that are obtained allow for the estimation of posteriordistributions via the Gibbs sampling algorithm. A toxicity data set involving the effect ofethylene glycol in mice is analyzed using our considered methodology.
Dissertation or thesis originally submitted to ProQuest
Appiah Kubi, John, "Bayesian and Maximum Likelihood Estimation for Correlated Binary data Based on the Double Binomial Distribution" (2021). Electronic Theses and Dissertations. 2878.
Available for download on Saturday, May 25, 2024
Data is provided by the student.