What question did this study set out to answer?

Develop and evaluate a computationally practical approximate inference framework for generalized linear mixed models using Laplace approximations, penalized quasilikelihood, and pseudo-likelihood.

March 1, 1993Journal of the American Statistical Association

Approximate Inference in Generalized Linear Mixed Models

View Full Paper

Ask AI

Bookmark

Authors

NBN. E. BreslowUniversity of Geneva DCDavid ClaytonUniversity of Southern California

Discussion

Member takes

Implication

Statistical modeling study reveals penalized quasilikelihood enables effective approximate inference across complex clustered data, suggesting broad utility despite slight bias in binary outcomes.

Key Points

Develop and evaluate a computationally practical approximate inference framework for generalized linear mixed models using Laplace approximations, penalized quasilikelihood, and pseudo-likelihood.
Approximated marginal quasi-likelihood functions via Laplace's method to construct penalized quasilikelihood (PQL) estimating equations for mean parameters and pseudo-likelihood for variance components.
Implemented estimation algorithms via repeated iterations of standard normal-theory restricted maximum likelihood (REML) procedures.
Evaluated performance using mathematical derivations, numerical simulations, and diverse empirical datasets spanning epidemiology, oncology, and experimental biology.
PQL delivers computationally feasible and reliable approximate inference for fixed effects and random-effects realizations across varied hierarchical data settings.
Estimates exhibit mild underestimation of variance components and absolute fixed effects for clustered binary data, though accuracy improves substantially when binomial denominators exceed one.

Cite This Study

Breslow et al. (1993) studied this question.

synapsesocial.com/papers/698ab37d1b45be50fa1f15b5 https://doi.org/10.1080/01621459.1993.10594284

View Full Paper

Ask AI

Bookmark

Implication

Statistical modeling study reveals penalized quasilikelihood enables effective approximate inference across complex clustered data, suggesting broad utility despite slight bias in binary outcomes.

Approximate Inference in Generalized Linear Mixed Models

Discussion

Member takes

Implication

Key Points

Cite This Study

Approximate Inference in Generalized Linear Mixed Models

Discussion

Member takes

Implication

Key Points

Cite This Study

Also Consider

Also Consider