A0715
Title: Beta regression models: Practical analyses with KNHANES 2013-2015 data and Covid-19 data
Authors: Jeong In Lee - Inha University (Korea, South) [presenting]
Seong Il Jo - Inha University (Korea, South)
Jae Oh Kim - Inha University (Korea, South)
Abstract: The continuous response variable with skewness and heteroscedasticity restricted to interval (0,1) violates traditional methods assumptions such as Ordinary Least Squares (OLS). The beta regression model is adequate for these situations. The beta distribution can be parametrized in terms of its mean and precision parameters, and the sub-model for mean can be estimated in beta regression. The variable dispersion beta regression is the extended beta regression model with two submodels for mean and precision. For these beta regression models, estimation can be performed by maximum likelihood or by Bayesian inference. Two practical applications are presented in comparison with traditional linear regression by OLS, beta regression and extended beta regression performed by ML and Bayesian inference. The first practical analysis is applied to Korean National Health and Nutrition Examination Survey (KNHANES) 2013-2015 to investigate the relationship between smoking and coffee intake. The second practical analysis is applied to the Covid-19 data set to examine the association between several county-level characteristics and the cumulative proportions of confirmed cases and deaths in the states of the USA. In these applications, it is illustrated that the beta regression and the extended beta regression are appropriate.
