Thomas Laitila: Finite Mixture of Tobit Models
Tid: On 2017-10-25 kl 13.00 - 14.00
Plats: Room B705, Department of Statistics, Stockholm University
Medverkande: Thomas Laitila, Örebro University
Karlsson and Laitila (2014) suggest a censored regression model estimator based on a finite mixture of Tobit models - the FMT estimator. This talk centres on the estimation of the FMT estimator covariance matrix using information-based estimators. The FMT estimator is defined, and related theoretical and empirical findings on covariance matrix estimation are presented and discussed. The conclusion arrived at: there is no obvious choice of covariance matrix estimator. Design of and results from a simulation study evaluating three estimators are presented and discussed. Our results suggest that the inverse of the negative Hessian matrix may be the overall best covariance matrix estimator for the FMT estimator. Concerning inference of parameters in mixture components, results are in favour of using the inverse of the outer-product of the scores matrix.
Reference:
Karlsson, M., Laitila, T. (2014). Finite mixture modeling of censored regression models. Statistical papers, 55(3): 627-642.