Remove cognitive overhead from your scientific papers

“There is this scientific convention of: ‘You put the images on one side, then you put the text to decipher it on the other side.’” That’s Jonathan Corum, science graphics editor for the New York Times, politely critiquing one of the ways in which a typical scientific paper creates unnecessary work for the reader, or “cognitive overhead.”

Decipher is the key word above (and a word I’ll use again below). If deciphering is necessary, it will precede understanding, but that doesn’t mean it is necessary. “No one intends to build a product with large cognitive overhead, but it happens if there isn’t forethought and recognition for it.”

A poorly designed figure. The reader’s eyes to have to dart back and forth between the panels, the colour bars, the labels, and the caption. This unnecessary work inhibits a deeper understanding of the data.

Corum’s statement about scientific figures is a special case of a rule articulated by Scott Berkun in his book How Design Makes the World: “the more instructions something has, the worse its design. It’s cheaper to add instructions later than to design something well”. People don’t want to read instructions before constructing their furniture or using their newest device. The same is true for scientific figures: what else are captions, but instructions? More generally, if you make good decisions in all aspects of producing your scientific paper, it will read less like a tedious instruction manual.

Don’t get me wrong; reading a scientific paper shouldn’t be easy. A reader won’t grasp abstract concepts or intricate theories without putting in some effort. A paper therefore presents a learning curve, but it should be one that meets the readers’s expectation. To quote Scott Berkun again: “I might be willing to suffer an intense two week learning curve to learn a new language, but I’d run screaming before investing that much time just to get $50 out of my local bank machine.”

Daniel Haight frames this same idea in statistical terms: “Clarity can be measured by how quickly you ‘get it’, normalized by information density.” High clarity is complexity easily grasped.  Low clarity is simplicity made difficult. This post is about removing low clarity elements from scientific papers.

Even minor mental obstacles add up

It’s not a stretch to imagine a scientific paper describing pizza preparation at a New York restaurant as follows: we placed the pizza in the oven at 21:56 UTC and removed it at 22:11 UTC. What’s important is that the pizza went in the oven around 6 pm and baked for 15 minutes. But to decipher that requires (i) converting from UTC (aka Greenwich Mean Time) to local time and (ii) subtracting the start time from the end time.

These bits of arithmetic are not difficult on their own, but they all add up. As do other minor mental obstacles like recalling what an newly invented acronym stands for, calculating a percentage change from raw data, or interpreting a line plot with a legend in the caption, not on the plot itself. If a reader has a finite capacity for attention, you don’t want them wasting it on these lesser details.

Data storyteller Elizabeth Ricks demonstrates the benefit of something as simple as annotating key data points. A lack of annotation “results in our audience having to do unnecessary work to understand our graphs”. In that same article, she shows how careful use of colour can “alleviate some of the mental effort our audience might encounter”.

You also don’t want to wast your own capacity for attention. Pick wisely when choosing what visualisation tools you use so that you become “engaged in the act of thinking, rather than distracted by the mechanics of using the software. Search out versatile scientific programming tools because “when you can do everything in the same language, you don’t have to suffer the constant cognitive switch costs: you can just keep solving the problem you’re trying to solve with as little cognitive overhead as possible”. If you’re still not convinced, consider web search. We take it for granted these days, but it was a novel idea 20 years ago that “if search engines were faster and better, they could be integrated into your thought process.”

Ways to reduce cognitive overhead

  • Remove unnecessary repetition. The date range 2012/06/24–2012/06/29, for example can be written as 24–29 Jun 2012. The former format forces us to compare both years and months only to find they’re the same.
  • Don’t try to be clever just to save a few words. Avoid the practice of sprinkling multiple bracketed statements to explain two contrasting cases simultaneously. That is, avoid the format of “Red (blue) shades denote positive (negative) values, which correspond to warmer (cooler) temperatures“. Also avoid “and/or”. Say “A, B, or both”. It’s not worth shortening to “A and/or B”.
  • Keep scientific quantities and their units together. In paragraphs, this is a no-brainer. In figures, however, issues creep in. The gif at the start, for example, labelled the quantities in the top left of each panel, included colour bars on the right of each panel, and relegated the units to the caption. (This poor design was inspired by a paper I recently read.)
  • Give datasets meaningful names. And recognise that a simple naming scheme might require more thought than you expect. In a recent paper, I compared oceanographic data across three days. An obvious, but lazy, way to refer to those throughout would have been Sep 27, Sep 29, and Oct 4. Or day 270, day 272, and day 277. Those dates are meaningful to me, but not the reader. Instead, I used the descriptive labels ‘calm’, ‘moderate’, and ‘windy’ throughout.
  • Forget the order in which you did the work. The example above generalises to the Methods section as a whole. To paraphrase from Stephen Heard’s guide to science writing: It’s tempting to write the Methods section chronologically, but your experience in doing the research doesn’t matter. What matters is your reader’s need to understand it.
  • Make use of preattentive attributes. Guide your reader’s eyes, before they even realise, to the most important parts of your figure with judicious use of colour, size, and position. For example, colouring a single bar in an otherwise grey bar chart asserts its importance.
  • Also make use of Gestalt laws of grouping to add order to complex arrangements of objects. Web designers follow these laws to make websites easy to navigate. Do the same for your scientific figures.
  • Look for chances to use pictures rather than words as a type of legend. The best example of this I’ve come across is the following from Baines (1995):
    In my own study of flow over an obstacle, I placed small pictures on the right to indicate the location of two-dimensional slices from a three-dimensional simulation:

Simplicity is appreciated, even if unrecognised

Cognitive overhead is like an uncomfortable piece of clothing: irritating when present, but unnoticed when absent. Readers may not recognise when a writer makes it easy for them. That said, readers may also not recognise that it is the writer’s fault for making things difficult. They’ll blame themselves instead.

Consciously or subconsciously, a reader burdened with cognitive overhead will vote with their eyes and attention, and go focus on something else. Which brings this post back to quoting Jonathan Corum: “I want to do the work. I don’t want my reader to do the work.”

Author: Ken Hughes

Post-doctoral research scientist in physical oceanography

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s