Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
167 tokens/sec
GPT-4o
7 tokens/sec
Gemini 2.5 Pro Pro
42 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
38 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Bayesian Inference from Composite Likelihoods, with an Application to Spatial Extremes (0911.5357v2)

Published 27 Nov 2009 in stat.ME, math.ST, and stat.TH

Abstract: Composite likelihoods are increasingly used in applications where the full likelihood is analytically unknown or computationally prohibitive. Although the maximum composite likelihood estimator has frequentist properties akin to those of the usual maximum likelihood estimator, Bayesian inference based on composite likelihoods has yet to be explored. In this paper we investigate the use of the Metropolis--Hastings algorithm to compute a pseudo-posterior distribution based on the composite likelihood. Two methodologies for adjusting the algorithm are presented and their performance on approximating the true posterior distribution is investigated using simulated data sets and real data on spatial extremes of rainfall.

Citations (181)

Summary

Bayesian Inference from Composite Likelihoods with Application to Spatial Extremes

The paper by Cooley, Davison, and Ribatet discusses a Bayesian framework for composite likelihood, a method increasingly utilized when full likelihood formulations are analytically intractable or computationally prohibitive. The authors move beyond frequentist properties associated with maximum composite likelihood estimators to establish a Bayesian machinery capable of integrating composite likelihoods into Bayes' theorem, thus proposing a structured approach for Bayesian inference when using composite likelihoods. They aim to ensure the propriety of the posterior density and propose modifications to correct for disparities between composite and full likelihood-derived posteriors.

Key Contributions

  • Proper Posterior Density: They assert that employing a composite likelihood results in a proper posterior density, a crucial aspect of Bayesian analysis.
  • Adjustments for Composite Likelihoods: The paper introduces two modifications—magnitude and curvature adjustments—to the composite likelihoods to mitigate differences in posteriors when compared to those obtained from the full likelihood. These adjustments are meant to allow credible intervals with adequate coverage, improving the utility of Bayesian inference using composite likelihoods.
  • MCMC Strategies and Theoretical Underpinnings: The research delineates how these adjustments can be incorporated into Metropolis-Hastings and Gibbs sampler algorithms, providing a robust theoretical foundation for their application. The aim is to yield credible intervals that are statistically valid, with coverage probabilities akin to results using full likelihoods.

Numerical Results and Empirical Discussion

Through simulation studies, the authors showcase the efficacy of these adjustments, particularly when dealing with Gaussian processes and hierarchical models for spatial extremes. They investigate how these adjustments influence posterior coverage, demonstrating that both magnitude and curvature adjustments approach performance levels seen with full likelihood-based posteriors but with more acknowledged uncertainty. Notably, this is showcased through a spatial rainfall dataset where the improved modeling of spatial dependence and marginal behavior is apparent.

Implications and Prospective Developments

This research presents substantial implications for the practical implementation of Bayesian models in spatial extremes, potentially transforming approaches that rely heavily on full likelihood formulations. The adjustments proposed not only improve computational efficiency but also facilitate more flexible model structures, yielding greater adaptability to complex real-world datasets.

In terms of broader theoretical perspectives, the exploration of adjusted composite likelihoods may inspire future developments in Bayesian analysis where the limitations of full likelihoods are prevalent. Moreover, extending these methodologies beyond spatial extremes could open avenues for novel applications in various domains where analytical complexity impedes likelihood formulations.

Conclusion

Cooley, Davison, and Ribatet's work represents a significant contribution to Bayesian inference methodologies, specifically in applications involving composite likelihoods and spatial extremes. It provides a well-founded framework for employing composite likelihoods in Bayesian models and opens doors for future explorations into more flexible, computationally feasible statistical procedures.