Funnel plots, outliers and smoking
I don't often blog the Journal of Public Health but a paper in the latest issue caught my eye - almost literally: there is an interesting graph of death rates from cancer in the 100 or so PCTs of England. The data are displayed as a funnel plot with age-standardised death rate on the 'y' axis and number of deaths on the 'x' axis. Doing it this way allows you to put a statistical outlier funnel on the map which narrows as the numbers get bigger.
In the first graph there is a group of PCTs which are outliers. Now comes the clever bit - if you take out lung cancer none of them are outliers. So these high death rate PCTs owe their unenviable position to smoking. The picture is somewhat different in females.
You could have done all of this with tables of data but it's the graphic which makes the story compelling.
While on the subject of images, here is a sorry tale. A coffee creamer, for reasons not entirely clear, uses on its packaging a graphic of a bear nursing a baby bear. All very cuddly but the effect of the picture is that 18% of women in Laos have used it as infant food. Analysis of images is known as semiotics, an important discipline because as this and other examples show, images often override words. (Which means you should go back to those cancer graphs and make sure the underlying analysis is fully rigorous.)