Joined: 13 Aug 2008
I would like to fit a bivariate spatial analysis with exponential covariance functions. I have succesfully got a univariate model working using mtrnv, but I'm not sure how to set up a bivariate model. In a previous post, in the context of an AR1 process Arthur has suggested something along the lines of ;
us(trait):mtrn(lat, long, phi=1)
However this fails with the error "ASReml failed to invert a spatially indexed correelation structure The data may have duplicate points or the correlations are too high."
In any case, I think that even if this could be made to work it would have the same phi for each trait whereas I would like separate phi's for each.
The bivariate model with separate phi's can be fitted in spBayes, but it does not allow me to have a cross-covariance function between the two traits with a seprate phi and this is critical to the analysis. Specifically I would like three correlation functions:
COR(y1_i, y_1j) = exp(-d/phi_1)
COR(y1_i, y_2j) = exp(-d/phi_12)
COR(y2_i, y_2j) = exp(-d/phi_2)
I was hoping that ASreml could be made to fit such a model.
Any indication of whether such a model would be possible would be great.