Title: Mixture of generalised linear models with unknown link
Authors: Sollie Millard - University of Pretoria (South Africa) [presenting]
Frans Kanfer - University of Pretoria (South Africa)
Mohammad Arashi - Ferdowsi University of Mashhad (Iran)
Abstract: An extension of mixtures of generalised linear models into a semi-parametric mixture setting is considered. The link function is estimated using a non-parametric estimation procedure. This approach allows for more flexibility since the non-parametric link function gives access to a larger subset of admissible distributions in the exponential family, whilst retaining much of the structure of a generalised linear model. Since the parameter estimates are not directly comparable to the traditional generalised linear model estimates, we consider using ratios of parameters for interpretation. A simulation study is used to evaluate the performance of the estimation procedure. The technique is illustrated using a toy example.