Abstract
Rice (1984) proposed a boundary modified kernel regression method which linearly combines two kernel regression estimators based on different bandwidths. In the context of density estimation, advantages of this method over two other popular approaches, local linear fitting and the boundary kernels of Müller (1991), are discussed. Selection of the ratio of the two bandwidths is studied. Asymptotic and exact mean squared errors are provided as tools to analyze the problem. In the case of Normal kernels, keeping the bandwidth ratio fixed, for ease and speed of implementation, and a specific bandwidth ratio are suggested.
| Original language | English |
|---|---|
| Pages (from-to) | 235–251 |
| Number of pages | 17 |
| Journal | Journal of the Chinese Statistical Association |
| Volume | 44 |
| Issue number | 3 |
| DOIs | |
| Publication status | Published - 1 Sept 2006 |
User-Defined Keywords
- boundary effect
- boundary kernel
- exact mean squared error
- kernel smoothing
- local linear smoothing