Dodgy data analysis has come under fire in recent months, and now an article in PLoS Biology1 takes aim at another staple of statistics. It argues that bar graphs used to describe a continuum of data are often uninformative and misleading, and should be purged from much of the scientific literature. The stance gained wide support from commenters on social media. Trevor Bedford, a virologist and computational biologist at the Fred Hutchinson Cancer Research Center in Seattle, Washington, tweeted:
Tom Oates, a kidney specialist at Imperial College Healthcare NHS Trust in London tweeted:
Researchers at the Mayo Clinic in Rochester, Minnesota, looked at more than 600 research articles published in top physiology journals in early 2014, and found that bar graphs were used to describe continuous data in more than 85% of the articles. This is problematic, the authors argue, because bar graphs that boil down data points to a single mean often fail to convey the nuances of the numbers. They add that very different data sets can be described by the same bar graph. “A visually appealing figure is of little value if it is not suitable for the type of data being presented,” they write. As an alternative, the authors recommend scatter plots that show every data point, especially for studies with relatively small sample sizes.
Bar graphs are designed to show categorized data. But they are often used for other types of data, partly because they are easier to make using Microsoft Excel than are other kinds of charts, according to the article. As a remedy, the authors provide free Excel templates for creating scatter plots. They also encourage journals to adopt policies to keep inappropriate bar graphs out of their pages. Finally, they suggest that investigators should receive extra training in data presentation.
Gaétan Burgio, a geneticist at the Australian National University in Canberra, was one of many researchers who shared the study on Twitter. He said in an interview that a lot of scientists, especially those in the biological sciences, seem to be unaware of the limitations of bar graphs. “They have become a standard that people use without questioning,” he says.
Bedford says that scatter plots showing every data point give readers a chance to interpret the data on their own. “Your eyes can do a lot to find patterns,” he says. “If there’s a story there, it should be obvious.”
- Journal name:
- Date published: