Over at Pharyngula PZ Myers looks at a series of graphs, and recommends using them. Zeno takes him to task over the poor quality of the data presentation. Having read Tufte, I am with Zeno on this. I’ve recently also done a tutorial with my clinical placement students on this subject, using what is one of the worst graphs I have ever come across. The problem is ratio. In a straight bar chart, increasing the height of the various bars shows a clear relationship between the values. Use a picture though and increasing the height increases the area, and you’re squaring the ratio; the visual impression of the graph is then skewed.

There are lots of things wrong with this graph, but let’s look at the golf tees. The ratio of the area of the tees distorts the visual impression, giving an illusion of a much greater difference between the minimum and maximum values (compounded by the break in the Y-axis). This effect is much worse because the golf tees do not represent any data at all. The data is represented by the golf balls. Delete the golf tees and you lose no data at all. Lose all the ink except the pixels at the centre of each of the golf balls and you don’t lose any information

Somewhere a doctor has found a programme that draws pretty pictures in graphs. He’s used it to show his skill in drawing pretty pictures – and at the same time has demonstrated that he’s completely unable to effectively display data.

Advertisements