ABSTRACT
Inference in problems in which the set of possible models is a parametric class, indexed by one or more parameters, is said to be parametric. In nonparametric inference, the basic problem may involve some distribution parameter such as the median or the mean, or some other descriptive quantity such as P(X <Y), but these would not characterize the population. The distinction between parametric and nonparametric is not always clear-cut, and there is not a universally accepted definition of “nonparametric.” Sometimes the contingency table analysis of Chapter 11 is included in the class of nonparametric methods. Most of the testing and estimation problems we’ve studied so far are parametric. In particular, estimating parameters is often parametric, and sounds like a parametric problem; but estimating a population median or the probability P(X < Y) is perhaps nonparametric.
