Simulated inference, with applications to wildlife population assessment

Large increases in computational power are leading to far-reaching changes in statistical inference. For statisticians, simulation used to be merely a tool for testing the validity of simple results from statistical inference when assumptions fail. Now, it is replacing traditional statistical inference. The practical use of simulated inference far surpasses the importance attached to it in undergraduate degree programmes. In this paper, we illustrate the power and flexibility of simulated inference, using examples from the field of wildlife population assessment. We restrict our attention to extensions of the bootstrap: incorporating model selection uncertainty into inference; use of a weighted bootstrap when there are insufficient data given the complexity of the model; bootstrapping spatial data in the presence of autocorrelation; and bootstrapping overdispersed counts.