Deviance information criteria for missing data models

By Gilles Celeux, Florence Forbes, Christian P. Robert, M. Titterington. 2006

943 citations recorded by the source corpus.

The deviance information criterion (DIC) introduced by Spiegelhalteret al. (2002) is directly inspired by linear and generalised linear models,but it is not so naturally de ned for missing data models. In this paper,we reassess the criterion for such models, testing the behaviour of variousextensions in the cases of mixture and random e ect models.

Subfields: Artificial Intelligence, Statistics and Probability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Statistics and Probability

Citations in the corpus

A Bayesian model for repeated measures zero-inflated count data with application to outpatient psychiatric service use (2010, methodology): “Note that this is a ‘conditional’ DIC akin to DIC #7 in Celeux et al. (2006).”
Bayesian Inference and Application of Robust Growth Curve Models Using Student's t Distribution (2013, background): “Especially, for models involving random-effects as in our robust growth curve models, the computation of DIC is still not very clear (Celeux et al., 2006).”
Variational approximations in Bayesian model selection for finite mixture distributions (2007, methodology): “2 Results of Analysis of Simulated Data from Mixtures of Univariate Gaussians The following example is taken from the paper by Celeux et al. (2006). Celeux et al. (2006) consider different ways of defining the DIC for model comparison in the case of latent variable models.”
Joint Bayesian model selection and parameter estimation of the generalized extreme value model with covariates using birth‐death Markov chain Monte Carlo (2009, background): “Celeux et al. [2006] discussed alternative representations of the DIC for latent variables.”
The General Projected Normal Distribution of Arbitrary Dimension: Modeling and Bayesian Inference (2017, background): “Several forms of DIC are available in the literature (Celeux et al., 2006).”
WinBUGS for population ecologists: bayesian modeling using markov chain Monte Carlo methods (2009, background): “However, the DIC statistic is in its infancy and is controversial (see the discussion papers following Spiegelhalter et al. 2002 and Celeux et al. 2006).”
Model Evaluation and Multiple Strategies in Cognitive Diagnosis: An Analysis of Fraction Subtraction Data (2008, methodology): “In addition, similar to the marginal likelihood, the conditional log-likelihood was computed by integrating out the incidental parameters, θ and α. Finally, the DIC was computed based on DIC4 for missing data models proposed by Celeux et al. (2006) .”
Estimating the Integrated Likelihood via Posterior Simulation Using the Harmonic Mean Identity (2006, background): “Also noteworthy is the forthcoming Bayesian Analysis discussion paper by Celeux et al. (2006), who offered a variety of " repaired " versions of pD for missing data (especially mixture model) settings.”
Understanding predictive information criteria for Bayesian models (2013, background): “Various diﬃculties have been noted with DIC (see Celeux et al., 2006, Plummer, 2008, and much of the discussion of Spiegelhalter et al., 2002) but there has been no consensus on an alternative.”
Bayesian Quantile Regression for Ordinal Models (2016, methodology): “In such a scenario, model selection criterion such as deviance information criterion or DIC (Spiegelhalter et al. 2002; Celeux et al. 2006) may be utilized to choose a value of p that is most consistent with the data.”
Regional fertility data analysis: A small area Bayesian approach (2013, positive): “To compare models and validate the importance of the spatial models, we use Deviance Information Criterion (DIC) for mixed-effects model, DIC4 (Celeux et al 2006) based on complete likelihood DIC4 = −4Eθα[log (Y α|θ)|Y ] + 2Eα[log (Y α|Eθ[θ|Y α])|Y ] , −4E1 + 2E2 where E2 = Eθ[θ|Y W ] can be…”
A survey of Bayesian predictive methods for model assessment, selection and comparison (2012, methodology): “Celeux et al. (2006) did not compare directly the bias or the variance of the different proposals, but tested them in a model selection with Galaxy data and one simulated data, and thus it is not easy to interpret the results with regard to ability to estimate the actual predictive performance as…”
Identification and Bayesian Estimation of Dynamic Factor Models (2015, methodology): “An alternative model comparison method is to use the deviance information criteria (DIC); see, for example, Berg, Meyer, and Yu (2004), Celeux et al. (2006), and Li, Zeng, and Yu (2013).”
The deviance information criterion: 12 years on (2014, positive): “As Celeux et al. (2006) showed well in the context of mixture models, DIC is not based on a universal principle that could lead to a procedure that was both computationally practical and generically applicable.”