Kernel Estimates for Categorical Data

From MaRDI portal





A new method of estimation of the smoothing parameters of the Aitchison- Aitken kernel method for categorical data [\textit{J. Aitchison} and \textit{C. G. G. Aitken}, Biometrika 63, 413-420 (1976; Zbl 0344.62035)] is proposed. This is based on minimizing an unbiased estimator of the mean squared error, and it provides a natural approach to multivariate estimation, thus allowing for the interaction structure of data in a simple way. The kernel smoothing technique is shown theoretically to offer a potential improvement over the direct use of cell frequencies, by means of an existence theorem akin to that used by Hoerl and Kennard for ridge regression. The proposed smoothing parameter estimates are compared with competing methods by means of a simulation in the simple binomial case and are illustrated on a set of multivariate data used in \textit{D. M. Titterington}, Technometrics 22, 259-268 (1980; Zbl 0441.62036). A property of invariance to the reversal of estimate formation and marginalization is proved for any multivariate kernel and estimation procedure.











This page was built for publication: Kernel Estimates for Categorical Data

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3750813)