Statistics Seminars: Conditional Inference in the Bootstrap Era
2 March 2007 14:00 in CM221
Among the most significant advances in statistical methodology over the last 25 years or so has been the development of highly accurate likelihood-based asymptotic methods of parametric inference. These have the key characteristic of being specifically constructed to respect the needs of conditional inference. This likelihood era has coincided with a bootstrap era which has seen introduction of simple, simulation-based methods which can yield very accurate inference, when viewed from a repeated sampling perspective, in many settings, especially in parametric problems. In this talk we examine the properties of unconditional bootstrap methods from the perspective of conditional inference, and reveal astonishing levels of conditional accuracy. We weigh the pros and cons of the two approaches to parametric conditional inference and suggest a winner.
Contact email@example.com for more information