The Intrinsic Bayes Factor for Model Selection and Prediction

By James O. Berger, Luis R. Pericchi. 1996

How researchers see this paper

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

Citations in the corpus

A Bayesian model for repeated measures zero-inflated count data with application to outpatient psychiatric service use (2010, methodology): “To overcome the computational difficulties of Bayes factors, several authors have recommended approximating the Bayes factor using a cross-validation estimate of the marginal likelihood f (y) (Berger and Pericchi, 1996; Gelfand and Dey, 1994; Gelfand, 1996).”
Expected‐posterior prior distributions for model selection (2002, methodology): “This prior also corresponds to the Intrinsic Prior resulting from the Arithmetic IBF (Berger and Pericchi, 1997).”
Objective Testing Procedures in Linear Models: Calibration of the p‐values (2006, background): “…of variable selection in regression Casella & Moreno (2006) argued that ‘since we are not confident about any given set of regressors as explanatory variables, little prior information on their regression coefficients could be expected’ so that an objective Bayesian analysis seems to be justified.”
Methods and Criteria for Model Selection (2004, background): “According to Berger and Pericchi (1996), who advocated this position, automatic methods are essential, because the statistician will often, at least initially, consider a wide range of models, for which it will not usually be feasible to specify all priors subjectively (on this point, see also Laud ”
The philosophy of Bayes’ factors and the quantification of statistical evidence (2016, methodology): “Variations on this basic approach, called ‘‘partial Bayes factors’’, have been suggested by multiple authors, including Aitkin (1991); Atkinson (1978); Berger and Pericchi (1996, 1998); Spiegelhalter and Smith (1982).”
Forecast Combination and Model Averaging Using Predictive Measures (2007, unknown): “The problem of selecting a particular training sample is avoided by averaging over all possible minimal subsets, as suggested by Berger and Pericchi ( 1996 ).”
Model averaging in ecology: a review of Bayesian, information-theoretic, and tactical approaches for predictive inference (2018, methodology): “One such correction is to calibrate the model on a part of the data first, use the result as new priors and then perform the analysis described above (intrinsic Bayes factor: Berger and Pericchi 1996, fractional Bayes factor: O’Hagan 1995).”
Bayesian predictions of low count time series (2005, unknown): “The posterior model probabilities, constructed via (2), with the added condition that P (M1|y) + P (M2|y) + P (M3|y) = 1, have arbitrary values as a consequence.”
The Prior Can Often Only Be Understood in the Context of the Likelihood (2017, background): “…(Jeffreys, 1961, Kass and Wasserman, 1996), long a popular default, is explicitly defined in terms of the likelihood function, and other suggestions such as fractional or intrinsic Bayes factors (O’Hagan, 1995, Berger and Pericchi, 1996) additionally depend on the concept of individual data points.”
Generalized Fiducial Inference: A Review and New Results (2016, background): “In fact, the fiducial factors are closely related to the intrinsic factors of Berger and Pericchi (1996, 2001).”
BAYES FACTORS AND MARGINAL DISTRIBUTIONS IN INVARIANT SITUATIONS (2016, background): “For discussions of the appealing properties of BI12, see Berger and Pericchi (1996a) .”
Comparing Asset Pricing Models (2017, methodology): “(9) Berger and Pericchi (1996) suggest averaging such results across all minimal training samples as a means of increasing stability of the procedure.”
Time for a change: a tutorial for comparing multiple classifiers through Bayesian analysis (2016, methodology): “This method is called Bayesian model comparison and uses so-called “Bayes factors” (Aitkin, 1991; Kass and Raftery, 1995; Berger and Pericchi, 1996).”
Properties of Bayes Factors Based on Test Statistics (2008, methodology): “Methods for overcoming the first problem were classified into three categories by Berger & Pericchi (1996).”