ABSTRACT
This chapter discusses nonparametric techniques. It provides flexible alternatives to the parametric techniques, they are useful in connection with graphical assessments of goodness of fit for complex models. Life tables have been used by demographers and actuaries for many years to describe and compare patterns of human mortality. A comment similar to that of the preceding paragraph applies to the Bayesian theory. T. S. Ferguson introduced Dirichlet processes, for the purpose of deriving nonparametric Bayesian estimates of a distribution function from uncensored data. Plots of the hazard or cumulative hazard function are often useful in assessing the fit of a parametric family of survival distributions to a given set of data. At the price of some rather artificial restrictions on the form of the survival distribution, we have been able to appeal to standard large-sample likelihood theory to justify the methods.
