Inaugural lecture at Imperial College London: The Surprising Conditional
Adventures of the Bootstrap
Professor Alastair Young (Mathematics, Imperial College London)
Monday 13 March 2006 18.00-19.00
Clore Lecture Theatre
Abstract
Many of the most significant advances in statistics over the last 50
years have been computational. Among these has been the development of
bootstrap methods of inference. These provide estimates of statistical
variability using empirical sampling models constructed from sample data
and simulation from this empirical model, to replace the analytic
calculations and approximations required by conventional approaches to
inference. The methods can provide astonishingly accurate estimates,
especially in the parametric context, when viewed from the perspective
of repeated sampling. But this perspective is narrow, and a more
appropriate viewpoint must take into account the demands of conditional
inference, which argues that to be relevant statistical inference should
be conditioned on certain features of the sample data being analysed.
Conditional bootstrap simulation is, however, generally awkward. In this
lecture we describe how particular bootstrap procedures, implemented
without regard to the appropriate conditioning, can yield inference
which respects that of an ideal conditional inference, to a surprisingly
high degree.
See also http://www.imperial.ac.uk/P7413.htm.
Details for getting to the Mathematics Department:
http://www.ma.ic.ac.uk/findus.php
Kind regards
Sofia
Dr Sofia Olhede
Lecturer in Statistics
Department of Mathematics
Imperial College London
SW7 2AZ London
UK
Tel:+44 (0) 20 7594 8568
http://www.ma.imperial.ac.uk/~sco
|