News (December 2010): Special issue of Statistica Sinica on composite likelihood methods, now published

Workshop "survivors" on the last day (bigger ¦ full-size)

Background

In many modern applications of statistical models, standard likelihood-based inference meets difficulties caused by high-dimensional interdependencies. Prominent application areas include the analysis of multivariate longitudinal and event-history data, spatial statistics, social network analysis, and bioinformatics. The problems encountered include prohibitively large computational demands (usually arising from the need to calculate integrals in many dimensions), and undue sensitivity to secondary modelling assumptions.

Various alternative approaches based on modification of the likelihood have been suggested in the research literature; composite likelihoods are instances of this, and they have been of rapidly increasing interest recently. Composite likelihoods are pseudo-likelihoods constructed by pooling likelihood components, with each component corresponding to a marginal or conditional event. A prominent special case is pairwise likelihood, based on components which are marginal likelihoods for pairs of observations.

The workshop

This international workshop reviewed the state of art of composite likelihood inference, and promoted vigorous discussion of foundations, applications and future developments. A further theme of discussion was the comparison of composite likelihood methods with alternative, computer-intensive approaches to inference in highly structured models, such as Markov chain Monte Carlo.

The workshop brought together active researchers in the theory and application of composite likelihood, for an intensive 3-day meeting at CRiSM, University of Warwick. A few key participants were specifically invited. An open call was made for abstracts from other potential contributors, from which the scientific committee selected (mainly on the basis of relevance to the workshop theme) for inclusion in the workshop programme. All invited and contributed talks were plenary, and in addition there was a lively poster session.

Invited speakers: M Aerts (Hasselt), P Fearnhead (Lancaster), N L Hjort (Oslo), H Joe (UBC), S Lele (Alberta), K-Y Liang (Johns Hopkins), B G Lindsay (Penn State), N Reid (Toronto), N Shephard (Oxford), P Song (Michigan), C Varin (Venice).

Scientific committee

Ruggero Bellio (Udine, Italy)
Paul Fearnhead (Lancaster, UK)
David Firth (Warwick, UK)
Nancy Reid (Toronto, Canada)
Cristiano Varin (Venice, Italy)
Grace Yun Yi (Waterloo, Canada)

Local organisers

David Firth (Warwick/NCRM/CRiSM; director)
Paula Matthews (Warwick/CRiSM; administrator)
Mohand Feddag (Warwick/NCRM)
Ioannis Kosmidis (Warwick/CRiSM)
Heather Turner (Warwick/NCRM)

Workshop funded jointly by ESRC (as an activity of the Lancaster-Warwick NCRM node) and by EPSRC (as an activity of CRiSM).

Institute of Mathematical Statistics


Royal Statistical Society (Research Section)