|
Abstract:
|
We propose a new class of spatial factor analysis. A number of factors are defined to drive the variability of the observations, as in standard factor models. Their loadings, however, are restricted to satisfy stochastic constraints based on their spatial structure. A few spatial structures can be considered, but we concentrate on those based on Gaussian processes---commonly used in geostatistics. Spatial dependence appears in the columns of the factor-loading matrix. A novel, reversible jump, Markov chain Monte Carlo algorithm is proposed to select the number of common factors. The new class of models is tested against synthetic and real data examples.
|
