Kernel-based density estimation using censored, truncated or grouped data

Abstract

Censoring, truncation and grouping represent different but related forms of incompleteness. Methods of producing kernel functions on the incomplete observations are proposed. They involve substituting for or averaging over the incomplete observations. Consistency of the procedures in terms of:the criterion of integrated mean squared error is established and optimal choice of smoothing parameter is achieved.

Keywords:

• incomplete data
• consistency
• smoothing parameter
• imputation