Imputation based on local likelihood density estimation for interval censored survival data with application to tree mortality in British Columbia

Censored data arise in many situations including forestry and medical studies, and may take several forms. In this project, we consider imputation methods for estimating lifetimes when interval censored data are available. We investigate an imputation method based on local likelihood density estimation, where kernel smoothing is used to estimate the underlying distribution of lifetimes in order to calculate the conditional expectation of the observed lifetime. We contrast this with a simple midpoint estimator, where the imputed lifetime is the midpoint of the interval censored data. We compare the two imputation methods in the context of an analysis of tree mortality in British Columbia. The main goal of the project is to describe the relationships between tree lifetimes and important covariates such as thinning levels and species of trees while observing how the use of different imputation methods can affect the derived relationships.