We model these random effects using a re-parametrisation of the Besag-York-Molliè conditional autoregressive prior distribution (Besag et al., 1991, Simpson et al., 2017).