An area-specific stick breaking process for spatial data
- 163 Downloads
Most of the existing Bayesian nonparametric models for spatial areal data assume that the neighborhood structures are known, however in practice this assumption may not hold. In this paper, we develop an area-specific stick breaking process for distributions of random effects with the spatially-dependent weights arising from the block averaging of underlying continuous surfaces. We show that this prior, which does not depend on specifying neighboring schemes, is noticeably flexible in effectively capturing heterogeneity in spatial dependency across areas. We illustrate the methodology with a dataset involving expenditure credit of 31 provinces of Iran.
KeywordsSpatial modeling Bayesian nonparametrics Stick-breaking process Geostatistical approach Fixed rank model Basis function
The editor-in-cheif and referees are gratefully acknowledged. Their valuable comments have improved the manuscript.
- Gelfand AE, Dey DK, Chang H (1992) Model determination using predictive distributions with implementation via sampling based methods (with discussion). In: Bernardo J et al (eds) Bayesian statistics. Oxford University Press, Oxford, pp 147–167Google Scholar
- Gelfand AE, Guindani M, Petrone S (2007) Bayesian nonparametric modeling for spatial data analysis using Dirichlet processes. In: Bernardo J et al (eds) Bayesian statistics. Oxford University Press, OxfordGoogle Scholar
- Gelfand AE, Banerjee S, Finley A (2013) Spatial design for knot selection in knotbased dimension reduction models. In: Mateu J, Muller W (eds) Spatio-temporal design: advances in efficient data acquisition. Wiley, Chichester, pp 142–169 (Chap. 7)Google Scholar
- Katzfuss M, Hammerling D (2014) Parallel inference for massive distributed spatial data using low-rank models. arXiv:1402.1472
- MacEachern SN (2000) Dependent Dirichlet processes. Technical Report, Department of Statistics, The Ohio State UniversityGoogle Scholar
- Petrone S, Guindani M, Gelfand AE (2008) Hybrid Dirichlet mixture models for functional data. Technical Report, Department of Statistics, Duke UniversityGoogle Scholar