Fig. 3

Using funnel plots to explore variability in hospital outcomes. Mortality rates (Y axis) are plotted against the annual number of eligible cases (X axis) for 40 hospitals with a horizontal line indicating the sample mean derived from the 40 observations and the inner and outer shaded areas indicating the 95% and 99% ranges respectively. In panel a (left) we vary the number of eligible cases from 0 to 4000 representing the range seen in our empirical data. In panel b (right) we randomly eliminate cases to reduce the sample size by 75% in each hospital to illustrate the effect on the 95% and 99% ranges. The reduction in case numbers available, as seen in the right panel, illustrates its effect on the potential for identification of outlying values and indicates that such analyses are likely to be most suitable for exploring variation across larger facilities or at sub-national regional levels