Abstract (endast på engelska):

Data collected from neighbouring districts are often spatially correlated. I will present the conditional  autoregressive (CAR) spatial model whereby these correlations can be fitted. The method has recently been implemented in the R package hglm and is described in an article to appear in the R Journal.

The hglm package fits linear mixed models  by iterating between generalized linear models. This algorithm will be presented and how  it can be used to fit a CAR model in a computationally efficient manner.

The package can also be used to make predictions for districts with no observations and an example using 4th grade school results from Ohio will be presented.