## Nonparametric Modelling of Hierarchically Exchangeable Data (2003)

### Cached

### Download Links

Citations: | 10 - 0 self |

### BibTeX

@MISC{Hoff03nonparametricmodelling,

author = {Peter D. Hoff},

title = {Nonparametric Modelling of Hierarchically Exchangeable Data},

year = {2003}

}

### OpenURL

### Abstract

Hierarchically exchangeable data are characterized by the exchangeability of a population of units and the exchangeability of observations from each individual unit. A flexible model for such data is the hierarchical logistic-normal model, which provides unconstrained sampling distributions at the within-unit level and an unconstrained covariance structure at the betweenunit level. Also, the sampling distribution at the between-unit level is unimodal in a weak sense. Parameter estimation and inference for the hierarchical logistic-normal model is relatively straightforward via Markov chain Monte Carlo or an approximate EM algorithm. These and other features of the hierarchical logistic normal model are explored, and the model is applied to the analysis of tumor locations in a mammalian population. A comparison is made to a similar data analysis based on Dirichlet distributions.

### Citations

799 |
Applied multivariate statistical analysis, p
- Johnson, Wichern
- 1998
(Show Context)
Citation Context ...re the conditional mode and negative inverse Hessian at the mode of f(zi|yi, µ l, Σl). The above is minimized by µ l+1 = 1 ∑m m i=1 ˆµ i and Σl+1 = 1 ∑m m i=1 [(ˆµ i − µ l+1)(ˆµ i − µ l+1) ′ + ˆ Σi] (=-=Johnson and Wichern, 1988-=-). Unrestricted estimates of Σ may be highly variable if m < |Y|. In this case, we may wish to fit Σ in terms of a smaller number of parameters, such as a diagonal matrix, or if there is a metric on Y... |

460 |
Mixtures of Dirichlet processes with applications to Bayesian nonparametric problems
- Antoniak
- 1974
(Show Context)
Citation Context ...(Y), K ∈ N, θk = 1 , k=1 k=1 where δf denotes a point mass on a distribution f. This latter set Q is in fact the class of distributions that results when one applies a mixture-of-Dirichlets approach (=-=Antoniak 1974-=-, Escobar 1995, West 1997, Escobar and West 1998) to this setting. Such an approach (which is discussed more in Section 6) may be appropriate if one is interested in identifying clusters of identical ... |

303 | Ferguson distributions via Pólya urn schemes - Blackwell, MacQueen - 1973 |

138 |
Exchangeability and Related Topics
- Aldous
- 1985
(Show Context)
Citation Context ...ent of Statistics, University of Washington, Seattle, Washington 98195-4322, U.S.A. This research was partially supported by NIH grant 1R01CA094212-01. Email: hoff@stat.washington.edu 1(Shaked 1977, =-=Aldous 1985-=-), and is also equivalent to assuming that observations within a unit are conditionally independent and identically distributed, given some unit-specific parameter. A common model for such structures ... |

122 | Nonparametric estimation of nonstationary spatial covariance structure - Sampson, Guttorp - 1992 |

72 | Unbalanced repeated-measures models with structured covariance matrices - JENNRICH, SCHLUCHTER - 1986 |

66 | Some aspects of Pólya tree distributions for statistical modeling
- Lavine
- 1992
(Show Context)
Citation Context ...utions, but perhaps inappropriate if one is interested in pooling information across similar yet distinct units. Another flexible hierarchical model could also be obtained using a Pólya tree process (=-=Lavine, 1992-=-; Mauldin et al., 1992), for which one can incorporate a certain amount of spatial dependence among {p(y), y ∈ Y}, depending on a choice of a partition tree of Y. The author has analyzed the data in S... |

44 | Modeling Household Purchase Behavior with Logistic Normal Regression - Allenby, Lenk - 1994 |

40 | Logistic-normal distributions: some properties and uses - Aitchison, Shen - 1980 |

37 | Density estimation, stochastic processes and prior information - LEONARD - 1978 |

28 | The Logistic Normal Distribution for Bayesian, Nonparametric, Predictive Densities - Lenk - 1988 |

26 | Computing nonparametric hierarchical models
- Escobar, West
- 1998
(Show Context)
Citation Context ...notes a point mass on a distribution f. This latter set Q is in fact the class of distributions that results when one applies a mixture-of-Dirichlets approach (Antoniak 1974, Escobar 1995, West 1997, =-=Escobar and West 1998-=-) to this setting. Such an approach (which is discussed more in Section 6) may be appropriate if one is interested in identifying clusters of identical units, as the discrete estimate of q models each... |

26 | Bayesian meta-analysis. In - DuMouchel - 1990 |

19 | Towards a practicable Bayesian nonparametric density estimator - Lenk - 1991 |

18 |
Hierarchical models for combining information and for meta-analysis. Bayesian Statistics 4
- Morris, Normand
- 1992
(Show Context)
Citation Context ... normal model (Hill 1965, Tiao and Tan 1965), in which the unit-level data and unit-specific parameters are both assumed to be normally distributed. More recent work on meta-analysis (Dumouchel 1990, =-=Morris and Normand 1992-=-) has also focused on parametric models for data and unit-specific parameters. Leonard (1977) proposed a model in which the data within each unit was modeled nonparametrically, in that the unit level ... |

16 |
Inference about variance components in the one-way model
- Hill
- 1965
(Show Context)
Citation Context ...ing that observations within a unit are conditionally independent and identically distributed, given some unit-specific parameter. A common model for such structures is the hierarchical normal model (=-=Hill 1965-=-, Tiao and Tan 1965), in which the unit-level data and unit-specific parameters are both assumed to be normally distributed. More recent work on meta-analysis (Dumouchel 1990, Morris and Normand 1992)... |

14 | Hierarchical generalized linear models and frailty models with Bayesian nonparametric mixing - Walker, Mallick - 1997 |

13 |
Data-Based Optimal Smoothing of Orthogonal Series Density Estimates
- Wahba
- 1981
(Show Context)
Citation Context ...egree of smoothness in our density estimates. With this in mind, another approach would be to model each unit-specific distribution p as having a Fourier-series expansion p(y) = 1 + ∑ k̸=0 zke 2πiky (=-=Wahba, 1981-=-). Variability among units could be modeled with a multivariate normal distribution on the Fourier coefficients zk. Although such an approach provides a flexible model for smooth densities, it is unap... |

11 | 1994) Inference for a covariance matrix - Brown, Le, et al. |

11 | A simulation-intensive approach for checking hierarchical models - Dey, Gelfand, et al. - 1998 |

10 | Embedding Nonnegative Definite Toeplitz Matrices in Nonnegative Definite Circulant Matrices, with Applications to Covariance Estimation - Dembo, Mallows, et al. - 1989 |

9 | A Bayesian method for histograms - Leonard - 1973 |

9 |
Bayesian analysis of random-effects models in the analysis of variance. I: Posterior distribution of variance components
- Tiao, Tan
- 1965
(Show Context)
Citation Context ...servations within a unit are conditionally independent and identically distributed, given some unit-specific parameter. A common model for such structures is the hierarchical normal model (Hill 1965, =-=Tiao and Tan 1965-=-), in which the unit-level data and unit-specific parameters are both assumed to be normally distributed. More recent work on meta-analysis (Dumouchel 1990, Morris and Normand 1992) has also focused o... |

8 |
A concept of positive dependence for exchangeable random variables
- Shaked
- 1977
(Show Context)
Citation Context ...nce ∗ Department of Statistics, University of Washington, Seattle, Washington 98195-4322, U.S.A. This research was partially supported by NIH grant 1R01CA094212-01. Email: hoff@stat.washington.edu 1(=-=Shaked 1977-=-, Aldous 1985), and is also equivalent to assuming that observations within a unit are conditionally independent and identically distributed, given some unit-specific parameter. A common model for suc... |

6 | Bayesiam simultaneous estimation for several multinomial distributions - Leonard - 1977 |

6 | A Bayesian Approach to Density Estimation - Thorburn - 1986 |

4 |
nonparametric Bayesian methods in hierarchical models
- Escobar
- 1995
(Show Context)
Citation Context ...= 1 , k=1 k=1 where δf denotes a point mass on a distribution f. This latter set Q is in fact the class of distributions that results when one applies a mixture-of-Dirichlets approach (Antoniak 1974, =-=Escobar 1995-=-, West 1997, Escobar and West 1998) to this setting. Such an approach (which is discussed more in Section 6) may be appropriate if one is interested in identifying clusters of identical units, as the ... |

3 | Bayesian Meta-Analysis. Statistical Methodology in the Pharmaceutical Sciences - DuMouchel - 1990 |

3 | Ferguson distributions via P6lya urn schemes. The Annals of Statistics - Blackwell, MacQueen - 1973 |

3 | Bayesian analysis of random-e¤ect models in the analysis of variance — I: Posterior distribution of variance-components - Tiao, Tan - 1965 |

2 |
Polya trees and random distributions’, The Annals of Statistics 20
- Mauldin, Sudderth, et al.
- 1992
(Show Context)
Citation Context ...rhaps inappropriate if one is interested in pooling information across similar yet distinct units. Another flexible hierarchical model could also be obtained using a Pólya tree process (Lavine, 1992; =-=Mauldin et al., 1992-=-), for which one can incorporate a certain amount of spatial dependence among {p(y), y ∈ Y}, depending on a choice of a partition tree of Y. The author has analyzed the data in Section 5 using a few s... |

1 | École d’été de probabilités de - SaintAllenby, Lenk - 1994 |

1 | Spectral representations for random densities - Lenk - 1987 |

1 | Bayesian predictive distributions under multinomial sampling,” in Bayesian and likelihood methods in statistics and econometrics: Essays in honor of George - J - 1990 |

1 | Hierarchical mixture models in neurological transmission analysis - unknown authors - 1997 |

1 | Excl1arlgE:abilil;y and related topies," in XIII-19B3, pp. 1198, ~pringier - t\l1 |

1 | Data-based optimal smoothing of orthogonal series density estimates - \Vahba - 1981 |

1 | Embedding nonnegative de Toeplitz matrices in nonnegative de circulant matrices, with application to covariance estimation - Dembo, Mallows, et al. - 1989 |

1 | Bayesian predictive distributions under multinomial sampling," in Bayesian and likelihood methods in statistics and econometrics: Essays in honor of George A - Lenk - 1990 |