Title: Semiparametric modelling in generalized linear models
Authors: Busayasachee Puang-Ngern - Macquarie University (Australia) [presenting]
Jun Ma - Macquarie University (Australia)
Ayse Bilgin - Macquarie University (Australia)
Timothy Kyng - Macquarie University (Australia)
Abstract: The semiparametric generalized linear models (SP-GLMs) are an extension of the well-known generalized linear models (GLMs). These semiparametric models are an alternative form of regression modelling. The additional nonparametric components in the conditional response density allow the distribution to be specified by the data while the response distribution is still in the exponential family. Iterative methods are applied to estimate the regression coefficient parameters and the nonparametric components simultaneously. We make a comparison of the biases, the asymptotic standard error, the Monte Carlo standard error, type I error and the likelihood ratio test of the regression coefficients between the SP-GLM and the GLM. Using simulation and a hypothetical parametric GLM to generate a sample of data, we fitted a SP-GLM to that data and found that the SP-GLM provides very similar results to the parametric GLM. Identical results can be found for logistic regression. The SP-GLM can provide more reasonable statistical inference for zero-inflated data.