Up is good: Avoiding confused physical metaphors when graphing data

In scientific research, the presentation of data, especially through graphical means, is pivotal in conveying complex information in an accessible manner. There is lots of guidance on how to do this well. One of the best examples is Leland Wilkinson’s book The Grammar of Graphics. But there are also some small and easily implemented pieces of advice. Here’s one such.

Woodin, Winter, and Padilla (2022) conducted a couple of experiments to test how easily people interpret graphs. In particular, they review the evidence that many cultures use a spatial or physical metaphor to represent the good-bad continuum. Good is up. Bad is down. For many people across the globe, this is a sort of unspoken assumption baked into the culture, in much the same way that Western cultures tend to represent time left-to-right rather than right-to-left.

The paper applies this physical metaphor to graphs and asks whether representing ‘bad’ with ‘up’ confuses people to some extent, because it runs against this subconscious association, learned from the culture, that ‘up’ is ‘good’.

If one is dealing with a negatively valenced construct — neuroticism or depression, say — then there are two obvious ways to graph it so that ‘good’ is still ‘up’. Either one inverts the construct (e.g. representing emotional stability instead of neuroticism, or happiness instead of depression) or one inverts the axes, which means the graph now shows as a deviation below the x axis. The more of the ‘bad’ construct, the further down it goes below the x axis:

The findings from their study reveal that non-inverted line graphs were generally easier to interpret than those with inverted axes, confirming the importance of adhering to conventional graphical standards. But they also found that graphs that adhere to the unspoken ‘good’ is ‘up’ convention were also more easily interpreted. And the effect sizes are not to sniff at. Errors in interpretation increase a few percent.

So what can we do with this information?

  1. Adhere to conventional graphical standards: Avoid inverting the axes of line graphs unless there is a very compelling reason to do so.
  2. Consider emotional valence in data presentation: When possible, frame data in positively valenced terms. For instance, instead of highlighting the rate of psychological inflexibility, consider framing the data around psychological flexibility. This can aid in more straightforward interpretation and reduce cognitive load on the audience.
  3. Don’t invert when it’s not valid to do so: Not all constructs have a well-defined opposite, and there are many ‘false friends’. For example, happiness is not the exact opposite of depression, but under some circumstances it might be OK to invert depression-like measures and refer to them as ‘well-being’. Each of us will need to make that decision on a case-by-case basis and balance issues around the accuracy of the construct label against interpretability.
  4. Apply more broadly: When graphing something, consider whether there are any likely physical metaphorical relationships with the construct or idea you’re representing, and try not to contradict these with your graph design. A really obvious example: for Western audiences, avoid representing time right-to-left.

Lastly, let me mention an anecdote. A few years ago I was working with some health professionals and used the phrase “less psychological flexibility correlates with….” I was trying to avoid talking about negative correlations, but I couldn’t get away from the double negative. And several people told me they found it difficult to follow the point. Many people find that double negatives trip them up. Talking about the opposites of established constructs might sometimes make this worse, or sometimes make it better. From my own experience, I can say pretty confidently that it’s usually easier for the audience if we talk about psychological flexibility rather than inflexibility.

If you know of other examples like this where experimental work leads to a simple and clear recommendation to improve science communication, please do let me know on twitter.