B0932
Title: Robust parameter estimation and variable selection in regression models when heteroscedasticity and skewness are present
Authors: Yesim Guney - Ankara University (Turkey)
Olcay Arslan - Ankara University (Turkey) [presenting]
Abstract: In many applications, not only the location of the response variables but also its scale and even skewness may depend on some explanatory variables. In these cases, modeling location, scale, and skewness may be needed to reflect all features of the data. The joint location, scale, and skewness model of the skew-normal distribution provide a useful tool for such responses where the normality assumption was relaxed to allow for skewness in the data. However, in the literature, the parameter estimation methods used for these models are typically limited to classical approaches which are sensitive to outliers. The other challenging problem for these models is selecting the important variables to estimate each parameter through an appropriate model. The maximum Lq-likelihood estimation method is first used to obtain robustness in estimating all model parameters, and then the penalized Lq-likelihood method is proposed to select the important variables in the three submodels. An expectation-maximization algorithm is implemented to obtain the parameter estimates and a simulation study and an application to real data is provided to demonstrate the performance of the proposed methods over the classical methods in the presence of outliers.
