Abstract
The necessity to replace smoothing approaches with a global amount of smoothing arises in a variety of situations such as effects with highly varying curvature or effects with discontinuities. We present an implementation of locally adaptive spline smoothing using a class of heavy-tailed shrinkage priors. These priors utilize scale mixtures of normals with locally varying exponential-gamma distributed variances for the differences of the P-spline coefficients. A fully Bayesian hierarchical structure is derived with inference about the posterior being based on Markov Chain Monte Carlo techniques. Three increasingly flexible and automatic approaches are introduced to estimate the spatially varying structure of the variances. In an extensive simulation study, the performance of our approach on a number of benchmark functions is shown to be at least equivalent, but mostly better than previous approaches and fits both functions of smoothly varying complexity and discontinuous functions well. Results from two applications also reflecting these two situations support the simulation results.
Item Type: | Paper |
---|---|
Keywords: | Gibbs sampling; Scale mixtures of normals; Spatially adaptive smoothing; Reversible Jump Markov Chain Monte Carlo |
Faculties: | Mathematics, Computer Science and Statistics > Statistics > Technical Reports |
URN: | urn:nbn:de:bvb:19-epub-2371-1 |
Language: | English |
Item ID: | 2371 |
Date Deposited: | 27. Mar 2008, 09:34 |
Last Modified: | 04. Nov 2020, 12:46 |