Rotation of parcels to form diamonds could avoid confusion between correlation and causality

The often repeated mantra that the correlation does not imply causation is not always effective. The media, the general public and even the scientists themselves are often guilty of seeing the cause-and-effect when there is only one association.

Two researchers now think they have a solution: when presenting the correlation results, simply tilt the 45-degree graph (see "Correlation or causal link").

The new graph, called diamond graph, is an original idea of ​​evolutionary biologist Carl Bergstrom and information scientist Jevin West of the University of Washington in Seattle. Together, both partners take a course on misuse of numbers, called Calling Bullshit.

They described the diamond plot in a document posted on the preprint server. arXiv last week¹. They provide a 45 ° rotation of scatter plots in a counter-clockwise direction, so that they become diamond-shaped with symmetrical axes. They hope that the unusual orientation of the graph will have a "shock" effect, forcing the viewer to explicitly examine – and hope to reject – the question of causality.

The researchers were inspired in part by the frequent misstatements of causality that they see in the popular press. "You can go through the last two weeks and find a lot of case examples," says Bergstrom. "Red wine is the cause, chocolate is the cause."

But Bergstrom adds that the media are not the only ones responsible: the scientists themselves are likely to make the same mistakes. In experimental methodology, scientists are taught that the independent variable – that controlled by the researcher – belongs to the x-axis of a graph, while the dependent variable – that which is measured – belongs to the 39, y-axis.

With respect to observational data, scientists can still apply these rules instinctively and mistakenly assume that the x-axis variable influences the other. Due to the symmetrical nature of the diamond pattern, neither axis takes precedence, which has reduced the risk of misallocation.

"I thought it was a good idea and I understood the logic of what they were doing," says Catey Bunce, a medical statistician at King's College London. "Do I think that using these plots will cause fewer people to confuse correlation and causality? Not at all.

According to Bunce, other factors may lead scientists to assume the cause, which could negatively impact the positive impact of diamond plots.

"I think advertising has something to do with it," she says. "If you write an article that says there is a causal link, it's a bold enough statement to make, and that will attract attention."

She adds that diamond plots, like any new statistical method, may have hidden dangers that only become apparent when people begin to misunderstand them.

Bergstrom recognizes that there is still some way to go before advocating that diamond plots become the norm. The team is now considering whether people can actually read diamond plots – and, if so, whether these plots reduce misinterpretation. "It helps or it does not work," he says. "At the moment, there is a proposal and the next step is of course to perform the appropriate user tests."