Fig. 4: Improvement in fit over time for the number of hospital deaths.

Each dot represents an analysis date (colour-coded) and region. For a data stream x_t and model replicates $y_t^i$ (where i accounts for sampling across the posterior parameter values) we compute the mean $\frac{1}{21}{\sum }_{T=t}^{t+20}{x}_T$; the prediction error $\frac{1}{21N}{\sum }_{i=1}^N{\sum }_{T=t}^{t+20}{\left(x_T-y_T^i\right)}^2$; the moving average $X_t=\frac{1}{6}\left(x_{t-3}+x_{t-2}+x_{t-1}+x_{t+1}+x_{t+2}+x_{t+3}\right)$; and the moving average error $\frac{1}{21}{\sum }_{T=t}^{t+20}{\left(x_T-X_T\right)}^2$. In each panel, the solid line corresponds to the path where the presented error statistic is equal to the mean.