Premium
Measuring the complexity of generalized linear hierarchical models
Author(s) -
Lu Haolan,
Hodges James S.,
Carlin Bradley P.
Publication year - 2007
Publication title -
canadian journal of statistics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.804
H-Index - 51
eISSN - 1708-945X
pISSN - 0319-5724
DOI - 10.1002/cjs.5550350108
Subject(s) - measure (data warehouse) , random effects model , bayesian probability , generalized linear model , mathematics , count data , hierarchical database model , hierarchical clustering , linear model , poisson distribution , statistical model , negative binomial distribution , bayesian hierarchical modeling , computer science , cluster analysis , statistics , bayesian inference , data mining , medicine , meta analysis
Measuring a statistical model's complexity is important for model criticism and comparison. However, it is unclear how to do this for hierarchical models due to uncertainty about how to count the random effects. The authors develop a complexity measure for generalized linear hierarchical models based on linear model theory. They demonstrate the new measure for binomial and Poisson observables modeled using various hierarchical structures, including a longitudinal model and an areal‐data model having both spatial clustering and pure heterogeneity random effects. They compare their new measure to a Bayesian index of model complexity, the effective number p D of parameters (Spiegelhalter, Best, Carlin & van der Linde 2002); the comparisons are made in the binomial and Poisson cases via simulation and two real data examples. The two measures are usually close, but differ markedly in some instances where p D is arguably inappropriate. Finally, the authors show how the new measure can be used to approach the difficult task of specifying prior distributions for variance components, and in the process cast further doubt on the commonly‐used vague inverse gamma prior.