## Bayesian P-Splines (2004)

Venue: | Journal of Computational and Graphical Statistics |

Citations: | 82 - 24 self |

### BibTeX

@ARTICLE{Lang04bayesianp-splines,

author = {Stefan Lang and Andreas Brezger},

title = {Bayesian P-Splines},

journal = {Journal of Computational and Graphical Statistics},

year = {2004},

volume = {13},

pages = {183--212}

}

### Years of Citing Articles

### OpenURL

### Abstract

P-splines are an attractive approach for modelling nonlinear smooth effects of covariates within the generalized additive and varying coefficient models framework. In this paper we propose a Bayesian version for P-splines and generalize the approach for one dimensional curves to two dimensional surface fitting for modelling interactions between metrical covariates. A Bayesian approach to P-splines has the advantage of allowing for simultaneous estimation of smooth functions and smoothing parameters. Moreover, it can easily be extended to more complex formulations, for example to mixed models with random effects for serially or spatially correlated response. Additionally, the assumption of constant smoothing parameters can be replaced by allowing the smoothing parameters to be locally adaptive. This is particularly useful in situations with changing curvature of the underlying smooth function or where the function is highly oscillating. Inference is fully Bayesian and uses recent MCMC techniques for drawing random samples from the posterior. In a couple of simulation studies the performance of Bayesian P-splines is studied and compared to other approaches in the literature. We illustrate the approach by a complex application on rents for flats in Munich.

### Citations

1612 | Generalized Additive Models
- Huang, Hastie, et al.
- 1990
(Show Context)
Citation Context ... sum of smooth functions f j . The response y is assumed to follow a Gaussian distribution y N(; 2 ). To allow for non-Gaussian responses the AM is extended to generalized additive models (GAM) (Has=-=tie and Tibshirani, 1-=-990) by assuming that the distribution of the response belongs to an exponential family and that the mean is related to the predictor by a response function h() = . A more general class of models inc... |

1252 | A Practical Guide to Splines - Boor - 1978 |

746 | Adapting to unknown smoothness via wavelet shrinkage - Donoho, Johnstone - 1995 |

512 | Bayesian analysis of binary and polychotomous response data - Albert, Chib - 1993 |

373 | On Gibbs Sampling for State Space Models - Carter, Kohn - 1994 |

329 |
Multivariate statistical modelling based on generelatized linear models
- Fahrmeir, Tutz
- 1994
(Show Context)
Citation Context ...models. Similar proposals have been made by Hastie and Tibshirani (2000) and Rue (2000). In generalized linear models parameter estimates are obtained by Fisher scoring or 12 in other words IWLS (see =-=Fahrmeir and Tutz, 2001-=-). In order to simulate from the posterior Gamerman suggests to combine IWLS and MH-updating of parameters. Suppose we want to update the regression coecientssj of the smooth function f j with current... |

267 | Bayesian image restoration with two applications in spatial statistics - BESAG, YORK, et al. - 1991 |

262 | Estimating optimal transformations for multiple regression and correlation - Breiman, Friedman - 1985 |

255 | Multivariate adaptive regression splines (with discussion - Friedman - 1991 |

215 | Varying-coefficient models - Hastie, Tibshirani |

156 | Polynomial splines and their tensor products in extended linear modeling (with discussion), The Annals of Statistics - Stone, Hansen, et al. - 1997 |

152 | Nonparametric regression using Bayesian variable selection - Smith, Kohn - 1996 |

148 | Efficient Simulation from the Multivariate Normal and Student-t Distributions Subject to Linear Constraints - Geweke - 1991 |

92 | Flexible parsimonious smoothing and additive modeling (with discussion - FRIEDMAN, SILVERMAN - 1989 |

91 | On conditional and intrinsic auto-regressions
- BESAG, KOOPERBERG
- 1995
(Show Context)
Citation Context ...the last step the main eects are centered in the same way as described before. Priors forsjs = (sjs11 ; : : : ;sjsmm ) 0 are based on spatial smoothness priors common in spatial statistics (see e.g. B=-=esag and Kooperberg, -=-1995). Since there is no natural ordering of parameters, priors have to be dened by specifying the conditional distributions ofsjs given neighbouring parameters and the variance component 2 js . The ... |

86 | The effect of improper priors on Gibbs sampling in hierarchical linear mixed models - Hobert, Casella - 1996 |

83 | Fast sampling of Gaussian Markov random fields
- Rue
- 2001
(Show Context)
Citation Context ...ditionals of the regression parameters. Drawing random numbers from the high dimensional distributions is, however, not trivial and is considerably facilitated by matrix operations for band matrices (=-=Rue, 2000-=-). Updating of the full conditionals in categorical probit models is facilitated by considering latent utility representations of such models. In this case, additional drawings from the latent utiliti... |

80 | 2001a): Bayesian Inference for generalized additive mixed models on Markov random field priors. Applied Stat - Fahrmeir, Lang |

77 | Efficient Sampling from the Posterior Distribution - Gamerman - 1997 |

67 | Hybrid adaptive splines - Luo, Wahba - 1997 |

56 | Computational Methods for Local Regression
- Cleveland, Loader
(Show Context)
Citation Context ...ared their Bayesian subset selection-based procedure with a variety of other approaches in the literature. In their simulation study they included MARS of Friedman (1991), Clive Loader's "loct&qu=-=ot; (see Cleveland and Grosse, 1991-=-), bivariate cubic thin plate splines with a single smoothing parameter, tensor product cubic smoothing splines withsve smoothing parameters, Breiman and Friedman's (1985) 17 a) global, 40 knots, j=3 ... |

51 |
Bayesian analysis of agricultural field experiments (with discussion
- Besag, Higdon
- 1999
(Show Context)
Citation Context ...ession parameters is replaced by a t-distribution. Such a prior has been already used in the context of dynamic models (Knorr-Held, 1996) and for edge preserving spatial smoothing (e.g. Higdon, 1994, =-=Besag and Higdon, 1999-=-). The classical GAM or VCM might be inappropriate for longitudinal data if heterogeneity among units or clusters is not suciently described by covariates. In that case, the introduction of an additio... |

41 | Direct generalized additive modelling with penalized likelihood - MARX, EILERS - 1998 |

38 | 2001b): Bayesian Semiparametric Regression Analysis of Multicategorical Time-Space Data - Fahrmeir, Lang |

36 | Flexible smoothing using B-splines and penalized likelihood - Eilers, Marx - 1996 |

35 | Conditional Prior Proposals in Dynamic Models
- Knorr-Held
- 1999
(Show Context)
Citation Context ...is paper are based on Albert and Chib (1993) and particularly on Fahrmeir and Lang (2001b). For general exponential family distributions we propose MH-algorithms based on conditional prior proposals (=-=Knorr-Held, 1999-=- and Fahrmeir and Lang, 2001a) and on iteratively weighted least squares proposals as suggested by Gamerman (1997) for generalized linear mixed models. Software forstting the models in this paper is i... |

32 | Generalized linear mixed models - Clayton - 1996 |

31 | Automatic Bayesian curve - MIYATA, Denison, et al. - 1998 |

24 | Bayesian analysis for correlated ordinal data models. In Generalized linear Models - A Bayesian Perspective (eds - Chen, Dey - 2000 |

23 | Adaptive Bayesian regression splines in semiparametric generalized linear models - Biller |

21 | Fitting multivariate regression functions by interaction spline models - Chen - 1993 |

20 | Computer Solution of Large Sparse Positive De Systems - George, Liu - 1981 |

19 |
Spatial Applications of Markov Chain Monte Carlo for Bayesian Inference
- Higdon
- 1994
(Show Context)
Citation Context ...r for the regression parameters is replaced by a t-distribution. Such a prior has been already used in the context of dynamic models (Knorr-Held, 1996) and for edge preserving spatial smoothing (e.g. =-=Higdon, 1994-=-, Besag and Higdon, 1999). The classical GAM or VCM might be inappropriate for longitudinal data if heterogeneity among units or clusters is not suciently described by covariates. In that case, the in... |

15 | A Bayesian Approach to Nonparametric Bivariate Regression - Smith, Kohn - 1997 |

13 |
Hierarchical Modelling of discrete longitudinal data, Herbert Utz
- Knorr-Held
- 1996
(Show Context)
Citation Context ...through a hierarchical t-formulation, i.e. the usual Gaussian prior for the regression parameters is replaced by a t-distribution. Such a prior has been already used in the context of dynamic models (=-=Knorr-Held, 1996-=-) and for edge preserving spatial smoothing (e.g. Higdon, 1994, Besag and Higdon, 1999). The classical GAM or VCM might be inappropriate for longitudinal data if heterogeneity among units or clusters ... |

12 | Function estimation with locally adaptive dynamic models - Lang, Fronk, et al. |

6 | Semiparametric Bayesian TimeSpace Analysis of Unemployment Duration - Fahrmeir, Lang, et al. - 2003 |

5 | Spatially adaptive penalties for spline fitting - Ruppert, Carroll - 2000 |

3 | Semiparametric generalized linear models: Bayesian approaches - Mallick, Denison, et al. - 2000 |

2 | Bayesian Generalized Additive Mixed Models. A Simulation Study - Lang, Fahrmeir - 2001 |

1 | The Computation of CGV Functions Through Householder Tridiagonalization With Application to the Fitting of Interaction Spline Models - Gu, Bates, et al. - 1989 |

1 | Bayesian Space-Time Analysis - Lang, Kragler, et al. - 2001 |