ruth@utstat.uucp (Ruth Croxford) (11/03/88)
Topic: Increasing (and decreasing) sensitivity of estimating functions Speaker: Don McLeish, University of Waterloo Date: Thursday, November 10, 4:00 p.m. Place: Room 2110, Sidney Smith Building, 100 St. George St., U of T Abstract: This talk will examine methods based on projections of controlling the sensitivity of estimating (or inference) functions to changes in the value of the underlying parameter. At the heart of the methods are two classes of functions, the insensitive (E-ancillary) and the sensitive (E-sufficient), analogous to ancillary and (complete) sufficient statistics. The method used is to project functions on the appropriate class (E-sufficient functions to increase sensitivity, E-ancillary to decrease, as when the parameter is a nuisance parameter). Various examples will be given indicating that this projection is a natural extension of the usual notion of marginalization, conditioning, and Rao- Blackwellization. Examples involving a mixture model will be developed, and other applications include robust estimates of location and a proportional hazards model. -------------------- Coffee and tea will be served in the Delury Lounge (SS6006) at 3:30 p.m.