Gibbs–type priors represent a natural generalization of the Dirichlet process: indeed, they select discrete distributions and can be characterized in terms of a system of predictive distributions that admit a simple and intuitive interpretation. By resorting to specific members of this wide class we address two important issues: (i) frequentist consistency; (ii) estimation of the discovery probability in species sampling problems. As for (i) we point out that Gibbs–type priors are consistent at discrete distributions whereas inconsistency may arise if the true distribution is diffuse thus highlighting that discrete nonparametric priors are typically consistent for the models they are designed for, namely data arising from discrete distributions. As for (ii), we emphasize how discrete nonparametric priors are well suited for addressing species sampling issues and display Bayesian nonparametric estimators of the probability of “discovering” a species that has appeared with any given frequency. These estimators can be compared with the frequentist counterparts, whenever the latter are available. Their practical use allows one appreciate the merits of the Bayesian nonparametric approach, which ensures that all objects of potential interest are modeled jointly and coherently.

Two tales about Bayesian nonparametric modeling

DE BLASI, Pierpaolo;FAVARO, STEFANO;PRUENSTER, Igor
2012

Abstract

Gibbs–type priors represent a natural generalization of the Dirichlet process: indeed, they select discrete distributions and can be characterized in terms of a system of predictive distributions that admit a simple and intuitive interpretation. By resorting to specific members of this wide class we address two important issues: (i) frequentist consistency; (ii) estimation of the discovery probability in species sampling problems. As for (i) we point out that Gibbs–type priors are consistent at discrete distributions whereas inconsistency may arise if the true distribution is diffuse thus highlighting that discrete nonparametric priors are typically consistent for the models they are designed for, namely data arising from discrete distributions. As for (ii), we emphasize how discrete nonparametric priors are well suited for addressing species sampling issues and display Bayesian nonparametric estimators of the probability of “discovering” a species that has appeared with any given frequency. These estimators can be compared with the frequentist counterparts, whenever the latter are available. Their practical use allows one appreciate the merits of the Bayesian nonparametric approach, which ensures that all objects of potential interest are modeled jointly and coherently.
Joint Statistical Meetings 2012
San Diego, CA, USA
28 Luglio 2012 - 2 Agosto 2012
JSM 2012 Proceedings
American Statistical Association
1696
1706
9780983937524
http://www.amstat.org/meetings/jsm/2012/
Bayesian nonparametrics; Bayesian consistency; Dirichlet process; discrete random probability measure; Gibbs–type priors; predictive distribution; species sampling
Pierpaolo De Blasi; Stefano Favaro; Antonio Lijoi; Ramses H. Mena; Igor Pruenster
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/2318/117814
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact