Chapter 8

Pages 217 and 229: From sp 1.1-0, overlay is defunct and over methods must be used. The code in geos_mod.R has been updated to correct the problem. Page 218: df constructed wrongly; "it would appear as though the estimate of the intercept is -2.47 and the estimate of beta1 is 6.95. I wonder if these labels are switched, as a simple lm of log(zinc) on sqrt(dist) has intercept 6.99 and slope -2.54 - there is a negative relationship of log zinc with sqrt(dist)"; thanks to Susan Service. Script updated 30 August 2017. Pages 219-221, section 8.5.10, chunk 82: Starting with gstat 1.1-0, the option to set cn_max is no longer needed (nor present from 1.1-1), due to the replacement of the meschach matrix library with the Lapack/BLAS library. The new behaviour in case of non-positive definiteness is printing of the warning (Covariance matrix singular at location [x,y,z]: skipping...) and continuing with the next location. Singularity is now by default detected by the Lapack routine dpotrs, which uses Choleski decomposition and breaks on non-positive definiteness. Gstat versions prior to 1.1-0 used the LDL' decomposition instead of the Choleski decomposition. Some non-positive definite matrices can be decomposed by LDL', but not by Choleski. Compatibility with the pre-gstat 1.1-0 behaviour is obtained when the variable choleski is set to the value 0, in the same way cn_max was defined.