Local linear hazard rate estimation and bandwidth selection

A new kernel based local linear estimate of the hazard rate, under the random right censorship model is proposed in this article. We study its finite sample and asymptotic properties and prove its asymptotic normality. Then we bring in three popular methods for bandwidth selection to the hazard setting as potential bandwidth choice rules for the estimate. We discuss their practical implementation and through Monte Carlo simulations we use four distributions with different hazard rate shapes to compare their performance over various sample sizes and levels of censoring.