Want to Make A Chart Memorable? Add Junk

A common criticism of charts is that they are filled with chart junk, and that removing the extraneous elements would make the chart better and stronger. That argument works for analytical charts, but not for charts that are used for presentation. The way memory works suggests that chart junk is actually very useful to not only get a point across, but make sure the reader remembers it.

Recap: The Chartjunk Paper

Bateman et al. looked into the effects of what is commonly referred to as chart junk: embellishments, graphical elements, etc. that are not strictly part of the chart, and that can be removed without reducing the amount of information the chart carries. The study found that the embellished charts were more memorable, in particular after a longer time period (two weeks) than the plain charts. Also, they found that embellishments did not interfere with subjects’ understanding of the data.

Criticism. Lots of It.

There has been considerable criticism of the paper, and much of it is probably valid: the sample size was small, the criteria left too much room for interpretation and/or experimenter influence, etc. Also, the charts used were very simple, with unstated assumptions about the data.

My point here is not to argue whether the criticism is justified, but that it misses the point. The bigger point we should get from the paper is that we need to be clearer about what we want the user to do with a chart.

Making Memories

We don’t remember isolated facts about the world. Confronted with information, we immediately try to make connections, and those connections help us remember facts. If you’re presented with some fact that has little or no connections with the other things you know about, you won’t remember it. A fact that fits well into the existing network of your knowledge will be retained much better.

We also embellish and elaborate information to remember it. That may be done subconsciously, or it may be an elaborate mnemonic device we use to make sure we retain a fact. Sometimes these techniques tie facts to certain locations, or we make up a little rhyme or poem around them. The interesting thing is that not only do these embellishments help us retain the information, in the long term we also remember the facts rather than the mnemonic devices.

The User and Her Task

There is a lot of talk about tasks in visualization, but most of the time, the task is analytical: analyze network traffic, find fraudulent transactions, etc. But what if the task is to remember a fact? Don’t we need to address that case differently than the usual analysis task?

The examples the Bateman study used were all taken from newspaper and magazine graphs that were not meant to be used for analysis. Rather, they presented information to the user, either as self-contained little information packages, or as illustrations accompanying articles.

The user task in these cases was not to analyze large amounts of streaming data or look for hidden patterns, but rather to take in a small chunk of information. The goal was not a new insight, but the communication of a fact.

What is so easily dismissed as chart junk therefore is actually a good idea in this case. In fact, the task requires something other than a boring, generic chart that looks like the next one. Unique embellishments help the user get the point, make connections, and remember the information. That is exactly what these charts were made to do, and the study showed that they did their job.

An interesting experiment would be to not only test people’s retention after two weeks, but after two months or a year. I bet the group that saw the embellished chart still had a memory of them, while the plain, boring charts are probably lost among all the other ones a person has seen in the meantime.

Thinking Outside the Analytical/Perceptual Box

For all its flaws, I am convinced that the Bateman et al. paper will prove to be highly influential in the longer run. But if it doesn’t, I think we’re in trouble.

Visualization has seen some research on perception, and that work has made its way into many techniques and rules we use today. We are still incredibly weak when it comes to higher-level cognitive processes like memory and learning, though. These processes are equally, if not more, important, in particular when it comes to presentation and communication of data. Data-to-ink ratio is a useful concept, but we need to start understanding when it should be used, and when other considerations are more important.

Published by

Robert Kosara

Robert Kosara is Senior Research Scientist at Tableau Software, and formerly Associate Professor of Computer Science. His research focus is the communication of data using visualization. In addition to blogging, Robert also runs and tweets.

12 thoughts on “Want to Make A Chart Memorable? Add Junk”

  1. Robert, I think it is important to distinguish between the chart junk studied in the Bateman paper (i.e. embellishments that could make the information more memorable) and your run-of-the-mill, Excel-style chart junk (e.g. extra grid lines, background colors, excessive use of colors). I agree that the former can help in this narrow area of communicating simple bits of information. The latter makes any chart harder to read and never adds value. For me, putting both things under the same heading of chartjunk confuses the issue.

  2. You’re right in principle, I was really only thinking about embellishments, not grid lines or other things. However, I can imagine there being a potential trade-off as well that might reduce contrast or otherwise interfere with chart reading a bit, but still help set a chart apart from the others. But I think in most cases, it should be very easy to figure out which chart junk is unequivocally bad, and which might help memory.

  3. I think (one of) the biggest problem(s) with this study is the simple fact that the ‘plain’ charts that are used for comparison are not well made charts.

    They do not follow *any* of the best practice rules for a good quality bar chart, and are not charts that anyone would seriously put forth as a good example of data visualization.

    They couldn’t help but fail.

  4. Another task that we could potentially seek to achieve is to motivate change. Incisive charts which just display facts are unlikely to achieve this objective. I have written earlier on the subject on my blog here – http://tinyurl.com/4yjm4zd

    This is a interesting and potential area of study – chart design based on what we are trying to achieve; presenting data that we would like to achieve, present data to motivate change, presenting analytical results, presenting financial or other similar data, providing an overview on a subject, mocking a certain point of view [ sarcastic comments as in the pie chart on gay marriage ]. All these categories may require different types of charts with difference in junk and other aspects. One rule cannot fit all. That is certain. Now we need to define the rule for the other categories of onjectices.

  5. How would the point be any less “got across” if you dispensed with any remaining graphiness, and just showed the headline, the detail text, and a picture of a monster?

  6. I think the difference is that you don’t always want to present the entire conclusion. The examples used in that study are a bit extreme in that way, the decorations could instead be based on the topic rather than the message. Also in either case, you’d still have some supporting evidence there, which might not only make the message stronger, but also more memorable as well (as opposed to being told “campaign costs are rising!”).

  7. Ah, so the graph here is used, as Andrew Lang is supposed to have said, “as a drunk uses a lamppost: more for support than illumination.”

    (the original quote is “he used statistics as a…”, but I can’t find out who the “he” being referred to was. The aphorism is often credited to Churchill instead of Lang, but I assume that’s just because Churchill is an aphorism credit magnet)

  8. Thanks, Robert, for your insightful revisitation of our paper. You have gotten to the crux of the matter in a way that many of the critiques have not: There is a place for non-data-ink in charts. We were not careful enough in the paper (although we do mention it) to differentiate between charts made for analysis and charts made for information presentation. The exemplar that inspired us was the USA Today graph, done much better by Holmes — something designed very much to tell a story. We admitted to ourselves that our charts in academic publications were trying to do exactly the same thing: convince our audience that our position, as described in the paper, was correct. So why were we relying on graphs and charts that would never, ever be used in publications that have this art down to a fine science?

    The key, then, was to determine whether properly designed charts that include chartjunk would interfere with comprehension — a significant barrier to using them in academic or otherwise informative publications. The answer was, more or less, ‘no’. Adding chartjunk does not really interfere with comprehension. It may interfere with analysis, but that’s not the purpose of these charts.

    The benefit, of course, is clear — we remember things with stories attached. If we create graphics with clear stories, emphasized with embellishments, people remember them.

    I should mention that this study was three years in the making. We tried numerous different versions and found little difference when we added non-data-ink that wasn’t clearly chartjunk. In other words, if we designed our plain graphs ‘better’ or even differently, we found no differences (much to our chagrin, since we had expected to find the data-ink ratio effect that Tufte hypothesizes in his work). It wasn’t until we used chartjunk that told a story, that emphasized the meaning of the graph, that we saw a significant difference in memorability.

    Incidentally… we never did see a difference in interpretability or comprehension, although we did time trials and error tracking on a variety of different comprehension and interpretation tests on a wide variety of charts intended to present information (not to support analysis). Of course, those are negative results and not easily publishable, but they’re of interest for chart designers, I would think.

    Thanks again!

Leave a Reply