Personal tools
You are here: Home › Examples › Spatial models › Geostatistics: Poisson GLMM
 

Geostistics: Poisson GLMM (irregular grid)

by Hans Julius Skaug — last modified Nov 28, 2011 07:48 AM
GLMM with spatial correlation, where the locations do not lie on a grid. Illustrates how you can parameterize a large correlation matrix in terms of an isotropic correlation function r(d), where "d" is the distiance between two locations.

Model description

Our data are 100 Poisson counts (y), each with parameter lambda. The datapoints are index by i and j (i,j=1,...,10). It is assumed that 

 

log(lambdai,j) = Xi,jb + ei,j.

 

where  Xi,jb is a linear predictor and ei,j are Gaussian random variables with covariance

cov(ei1,j1,ei2,j2) = s2 exp(a-1 d),

 

Here d is the Euclidean distance between the two positions.

 

Files

See "Navigation" box to the left.

  • .tpl:  Model file
  • .dat: Data file
  • .pin: Starting values for the numerical optimizer
  • .par: Result file (what you get when you compile and run your model)

 

 

Document Actions