CMStatistics 2017: Start Registration
View Submission - CMStatistics
Title: Sampling of pairs in composite likelihood estimation for latent variable models for categorical responses Authors:  Irini Moustaki - London School of Economics (United Kingdom) [presenting]
Ioulia Papageorgiou - Athens University of Economics and Business (Greece)
Abstract: Pairwise likelihood estimation has been recently developed for estimating the parameters of latent variable and structural equation models. Pairwise likelihood is a special case of composite likelihood methods that use lower order conditional or marginal log likelihoods. The composite likelihood to be maximised is a weighted sum of marginal or conditional loglikelihoods. Weights can be chosen to be equal or unequal for increasing efficiency. We approach the problem of weights from a sampling perspective. More specifically, we propose a sampling method for selecting pairs that is based on those pairs that contribute more to the total variance from all pairs. We demonstrate the performance of our methodology using simulated data and a real example.