| dc.description.abstract | Logistic regression is commonly used in research to assess the relationship of proportion with one or many variables. In logistic regression, when variance of a binomial response variable is larger than it should be (overdispersion), either the model or the parameter estimation needs to be modified. An alternative that can be applied is beta-binomial regression. Parameter estimation for logistic and betabinomial regression generally done by maximizing the likelihood function through the Iteratively Reweighted Reweighted Least Square (IRLS) algorithm. However, this algorithm requires much auxiliary information to work properly such as initial domain and differential. This study is purposed to examine the application of genetic algorithm as an alternative method for estimating logistic and beta-binomial regression parameters. The result shows that genetic algorithm can generate solutions that are close to IRLS even with better log-likelihood value. | en |
