BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Designing efficient composite likelihoods - Cristiano Varin (Unive
 rsità Cà Foscari di Venezia)
DTSTART:20170704T131500Z
DTEND:20170704T140000Z
UID:TALK73143@talks.cam.ac.uk
CONTACT:INI IT
DESCRIPTION:Composite likelihood is an inference function constructed by c
 ompounding component likelihoods based on low dimensional marginal or cond
 itional distributions. Since the components are multiplied as if they were
  independent\, the composite likelihood inherits the properties of likelih
 ood inference from a misspecified model. The virtue of composite likelihoo
 d inference is &ldquo\;combining the advantages of likelihood with computa
 tional feasibility&rdquo\; (Reid\, 2013). Given the wide applicability\, c
 omposite likelihoods are attracting interest as scalable surrogate for int
 ractable likelihoods. Despite the promise\, application of composite likel
 ihood is still limited by some theoretical and computational issues which 
 have received only partial or initial responses. Open theoretical question
 s concern characterization of general model conditions assuring validity o
 f composite likelihood inference\, optimal selection of component likeliho
 ods and precise evaluation of estimation uncertainty. Computational issues
  concern how to design composite likelihood methods to balance statistical
  efficiency and computational efficiency.<br><br>In this talk\, after a cr
 itical review of composite likelihood theory\, I shall focus on the potent
 ial merits of composite likelihood inference in modeling temporal and spat
 ial variation of disease incidence. The talk is based on past work with Na
 ncy Reid (Toronto) and David Firth (Warwick)\, and various new&nbsp\;proje
 cts with Manuela Cattelan (Padova)\, Xanthi Pedeli (Venice) and Guido Masa
 rotto (Padova). &nbsp\;
LOCATION:Seminar Room 1\, Newton Institute
END:VEVENT
END:VCALENDAR