|Publication details||In "COMPSTAT 1998 Proceedings", 203-208. Eds. Payne, R. and Green, P.. Physica-Verlag, Heidelberg.|
|Publisher details||Physica-Verlag, Heidelberg|
|Keywords||Kernel density estimation, Markov chain Monte Carlo, cross-validation|
A new procedure is proposed for bandwidth selection in univariate kernel density estimation. Rather than concentrate on minimising some criterion based upon the mean integrated square error (MISE), which depends directly on the (unknown) true density, we build a model for the data and use sampling methods to make inferences about the bandwidths. The model is Bayesian, and it is noted that it allows for systematic adjustment for subjective changes in smoothness of the density estimate.