## Avoiding model selection in Bayesian social research (1994)

Venue: | Sociological Methodology |

Citations: | 3 - 1 self |

### BibTeX

@ARTICLE{Gelman94avoidingmodel,

author = {Andrew Gelman and Donald B. Rubin},

title = {Avoiding model selection in Bayesian social research},

journal = {Sociological Methodology},

year = {1994},

volume = {25}

}

### OpenURL

### Abstract

Introduction Raftery's paper addresses two important problems in the statistical analysis of social science data: (1) choosing an appropriate model when so much data are available that standard P-values reject all parsimonious models; and (2) making estimates and predictions when there are not enough data available to fit the desired model using standard techniques. For both problems, we agree with Raftery that classical frequentist methods fail and that Raftery's suggested methods based on BIC can point in better directions. Nevertheless, we disagree with his solutions because, in principle, they are still directed off-target and only by serendipity manage to hit the target in special circumstances. Our primary criticisms of Raftery's proposals are that (1) he promises the impossible: the selection of a model that is adequate for specific purposes without consideration of those purposes; and (2) he uses the same limited tool for model averaging as for model selection, thereby

### Citations

1250 | Bayesian Data Analysis
- Gelman, Carlin, et al.
- 1995
(Show Context)
Citation Context ...cisms about his presentation of BIC as "the Bayesian approach to hypothesis testing, model selection, and accounting for model uncertainty." The Bayesian approach is a general one, which we =-=advocate (Gelman, Carlin, Stern, and Rubin, 1995-=-), and it is important to recognize that there is no single Bayesian solution to a statistical problem. Bayesian approaches to the problems posed by multiple models include exact Bayes factors using p... |

265 |
Estimation with quadratic loss
- James, Stein
- 1961
(Show Context)
Citation Context ...known that Bayesian approaches, which assign a prior distribution to the parameters in the model, can yield parameter estimates and predictions that are better from the frequentist perspective (e.g., =-=James and Stein, 1960, Efron an-=-d Morris, 1971, 1972). Different estimation procedures correspond to different prior distributions; for example, "ridge regression" corresponds to a normal prior distribution on the coeffici... |

113 |
Bayesianly justifiable and relevant frequency calculations for the applied statistician
- Rubin, B
- 1984
(Show Context)
Citation Context ...some examples in educational research); and posterior predictive checks, in which models are compared not by posterior probabilities but rather by their predictive accuracy for intended purposes (see =-=Rubin, 1984-=-, and Gelman, Meng, and Stern, 1995). Moreover, BIC cannot be construed as an approximation to any exact Bayesian solution, even a Bayes factor. In models with improper prior distributions (which incl... |

53 | Why are American presidential election campaign polls so variable when votes are so predictable - Gelman, King - 1993 |

36 |
Estimation in parallel randomized experiments
- Rubin
- 1981
(Show Context)
Citation Context ...ific interest. More generally, deviations from a model can be compared to their posterior predictive distribution, a Bayesian generalization of the reference distribution used for classical P-values (=-=Rubin, 1981-=-, 1984). Here, a Bayesian analysis of a posited model (e.g., quasi-symmetry) is used to generate hypothetical replicates of the data under their posterior predictive distribution. If the replicates ar... |

17 | Forecasting the Presidential Vote in the States - Campbell - 1992 |

14 | Testing in latent class models using a posterior predictive check distribution - Rubin, Stern - 1994 |

13 |
Using empirical Bayes techniques in the law school validity studies
- Rubin
- 1980
(Show Context)
Citation Context ... analysis of this example. With several years of data, regression coefficients can be pooled or partially pooled across years (in the same way that coefficients are partially pooled across schools in =-=Rubin, 1980-=-) using Bayesian methods. Other useful steps would be disaggregating the data (e.g., by race, sex, and age) and building an appropriate hierarchical model. Certainly, whether or not this extra informa... |

10 |
Limiting the risk of Bayes and empirical Bayes estimatorsâ€”Part II: The empirical Bayes case
- Efron, Morris
- 1972
(Show Context)
Citation Context ...roaches, which assign a prior distribution to the parameters in the model, can yield parameter estimates and predictions that are better from the frequentist perspective (e.g., James and Stein, 1960, =-=Efron and Morris, 1971, 1972). D-=-ifferent estimation procedures correspond to different prior distributions; for example, "ridge regression" corresponds to a normal prior distribution on the coefficients in a regression mod... |

9 | Bayesian regression with parametric models for heteroscedasticity - BOSCARDIN, GELMAN - 1996 |

7 |
Bayesian model checking using tail area probability", Statistica Sinica (with discussion
- Gelman, Meng, et al.
- 1995
(Show Context)
Citation Context ...educational research); and posterior predictive checks, in which models are compared not by posterior probabilities but rather by their predictive accuracy for intended purposes (see Rubin, 1984, and =-=Gelman, Meng, and Stern, 1995-=-). Moreover, BIC cannot be construed as an approximation to any exact Bayesian solution, even a Bayes factor. In models with improper prior distributions (which include all the examples in Raftery's p... |