I’m using GLIMMIX to run a generalized linear model with a binary response and two covariates. I’m also using the option rsmooth, the low-rank radial smoother, to model spatial correlation in 2-dimensions. Because the random effect variance is close to zero, I needed to rescale my spatial coordinates to achieve convergence (Schabenberger and Gotway, 2005 p.330). The problem I’m having is that parameter estimates (particularly the intercept) change as I change the factor I use to rescale the coordinates.
Here’s my code. Any thoughts? Thank you.
proc glimmix data = burndataxyCONIF1415_TOPcat3000S;
model GRID_CODE(order=freq ref=first) = ELEV_M SLOPE_DEG/solution ddfm=residual
link = logit;
random Xscale Yscale /type=rsmooth;