Composite marginal likelihood inference

 

In a number of applications, the presence of large sets of correlated data or the specification of complex models make unfeasible the use of the likelihood function, since too computationally demanding. One possibility is to avoid ordinary likelihood methods, or Bayesian strategies, and to adopt simpler pseudolikelihoods, like those belonging to the composite likelihood class. A composite likelihood consists in a combination of valid likelihood objects typically related to small subsets of data. It has good theoretical properties and it behaves well in many challenging applications. Examples include spatial statistics, multivariate survival analysis, generalized linear mixed models, frailty models, genetics.

In this talk, I will discuss the main properties of composite marginal likelihoods, like the pairwise likelihood and its extensions. Time series applications will be considered.