ABSTRACT
This chapter shows that the bootstrap adjusted estimator should also be bootstrapped. However, the method has been generalized much further to attack problems like discrimination, where objects are sorted into different classes, and certain problems for stochastic processes where the data are highly dependent. In fact, like most good ideas it is not very complicated, and the method might well have come up in the early days of statistics if it had not been so dependent on computer simulations. But the selection of this particular method among all possible ways to use the computer, and the recognition of its capacity is built on a very deep feeling for statistical inference. The original bootstrap is distribution free, which means that it is not dependent on a particular class of distributions. By this technique one can approximate the distributions of the parameter estimates for many well known distributions such as the Poisson, Binomial, and Exponential.
