BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Working with over-parameterized models - David Firth\, University 
 of Warwick
DTSTART:20090612T150000Z
DTEND:20090612T160000Z
UID:TALK18121@talks.cam.ac.uk
CONTACT:8047
DESCRIPTION:A parametric representation of a statistical model may\ninvolv
 e some redundancy\; that is\, the mapping from\nparameter space to family 
 of distributions may be\nmany-to-one.  Such over-parameterized representat
 ions are\noften very useful conceptually\, but can cause computational\nan
 d inferential problems (ridges in the likelihood\,\nnon-estimable paramete
 r combinations).  For linear and\ngeneralized-linear models\, well known a
 pproaches use either\na reduced basis or a generalized matrix inverse.  In
  this\ntalk I will discuss how to work with over-parameterized\nnonlinear 
 models.  Aspects covered will include\nmaximum-likelihood computation\, de
 tection of\nnon-identifiability\, and presentation of results.  Some\nimpl
 ications for Bayesian analysis will also be touched\nupon.  The work is mo
 tivated by the design and\nimplementation of the R package "gnm" (written 
 jointly with\nDr Heather Turner).\n
LOCATION:MR12\, CMS\, Wilberforce Road\, Cambridge\, CB3 0WB
END:VEVENT
END:VCALENDAR