|
|
Conditional Properties of Unconditional Parametric Bootstrap
|
When
|
2.00pm on Friday 24th November
|
|
Where
|
B617, Leverhulme Library, Columbia House
|
|
Presentations
|
|
|
Speaker
|
Alastair Young
|
|
From
|
Imperial College London
|
|
Abstract
|
Parametric bootstrap procedures are known to yield highly accurate inference on a scalar parameter of interest, when viewed from a repeated sampling perspective. But, little is known about the properties of bootstrap methods from the perspective of conditional inference. In this talk, we focus on a parametric bootstrap procedure, in which the distribution of the signed root likelihood ratio statistic is simulated, under the model in which a nuisance parameter is specified as its constrained MLE for a given value of the interest parameter. The conditional properties of the procedure are described for two contexts: multiparameter exponential families, where conditioning is used to eliminate a nuisance parameter, and models which admit an ancillary statistic, where conditioning is used to ensure relevance to sample data. The prescription, which ignores conditioning, is seen to yield inference in both contexts which respects conditioning to a remarkably high degree.
|
|
For further information
|
Thomas Hewlett (Postgraduate Administrator) Ext. 6879
Department of Statistics, Columbia House
|
|
|