BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Focussed Information Criteria for Model Selection and Model Averag
 ing - Nils Hjort\, University of Oslo
DTSTART:20111021T150000Z
DTEND:20111021T160000Z
UID:TALK34059@talks.cam.ac.uk
CONTACT:Richard Samworth
DESCRIPTION:The traditional approaches to model selection\, e.g. those bas
 ed on the Akaike and Bayesian information criteria (AIC and BIC)\,\nwork i
 n "overall modus"\, without considering what the selected\nmodel actually 
 may be used for later in the inference process.\nI shall discuss various v
 ersions of focussed information criteria\n(FIC) for different types of sit
 uations\, where the operating idea\nis to take explicitly on board what th
 e focus of the analysis is.\nThus I decide to not see it as particularly c
 ontradictory that\none model may be best for analysing say the mean struct
 ure\nwhereas another model may be better for analysing say the\nskewness s
 tructure (with the same set of data and the same list\nof candidate models
 ). I will first review the basic FIC machinery\ndeveloped in joint earlier
  work with Gerda Claeskens (cf. several\nJASA papers and our 2008 CUP book
 ) for the case of comparing\n(and averaging over) a class of parametric ca
 ndidate models\nand then present some ongoing work with one of my Oslo stu
 dents\,\npertaining to FIC comparisons between parametric and\nnonparametr
 ic models. Such problems are of a different\ncharacter in that one needs t
 o compare models with likelihoods\nwith models without likelihoods.\n
LOCATION:MR12\, CMS\, Wilberforce Road\, Cambridge\, CB3 0WB
END:VEVENT
END:VCALENDAR