Correspondence (letter to the editor): Statistics Visualization not Very Successful

In their article, Melchior et al. attempt to convey different percentages graphically using a modified pie chart (1).

The study cohort comprised 567 191 pregnant women. In the outer part of the pie diagram, the proportion of pregnant women is shown with respect to screening implementation. For example, only the pre-test was used in 63.3% (approximately 359 032) of the pregnant women. The prevalence of gestational diabetes (GDM) was 13.2% (approximately 74 869 pregnant women). Of these 13.2%, 4.4%—that is, around 24 956 pregnant women (and not 4.4% of 13.2%, which would be 3294 pregnant women)—fell into the “only pre-test” category. This means that approximately 33% of all GDM pregnant women were examined only with the pre-test. Thus, the pie slice should have been 120 degrees, since in pie charts, the sum of all pie slices must be 360 degrees. Instead, the authors used an angle of 15.8 degrees (15.8/360 degrees = 4.4%).

Even if the authors intended to deviate from the usual pie charts, it remains difficult due to the physiology of the human eye: 1) to correctly recognize angles at all, and 2) to correctly recognize angles of pie slices that are not contiguously distributed over 360 degrees. If the authors think this is irrelevant as the percentages are given over the pie slices anyway, the question then arises about the value of the diagram. In general, pie charts are viewed critically. Indeed, Cleveland and McGill pointed out already in 1985 that pie charts, which require angle degrees to be recognized by eye, are interpreted less exactly than bar charts, which require only heights or lengths to be determined by eye (2).