Hi Steve,

I’ve been following your blog for a few months now (since I first saw it), and I thoroughly enjoy it, have learned some things from it, and by and large agree with you about the problems with information delivery and visualization.

So I hope you don’t take any offense that I’m not completely on page with you in this instance. :-)

I actually like the Guardian’s chart and in many ways prefer it to the bar charts. The first reason you’ve admitted — it’s more engaging on a purely visual level, however I feel that you undervalue that aspect. Journalism, I believe, should have a portion of responsibility to engage, and while that should never trump truth and clarity (else you get pandering or outright lying), I think it’s important. The Guardian’s chart certainly contains no mis-truths, and I think its clarity is more open for debate.

As a first example, the color use in the Guardian’s chart is near 100% coverage of the overall page. While it may be true that comparing areas for nearly-similar values is difficult, it’s made easier by this large coverage area. The bar charts, in comparison, take up a very small portion of the page.

That becomes vitally important in the colors of smaller values. For example, I cannot tell in your bar chart the difference between the “Roman Abramovich” and “Aid” bar’s colors nearly as readily as I can in the original.

Also, the larger color coverage allows me to more quickly search for colors. I can identify very quickly which boxes are “Fighting” and which are “Giving”, but a scan of the bar chart does not make that as easy.

Nextly, McCandless’ chart includes some nesting (Tesco Profits, Tax Fraud, NHS IT spending), which is completely lost in the bar chart — I’ll not argue on whether it’s necessary, but it’s certainly information and is very clear from the nested boxes.

Last, back to the area vs. length argument, you disapprove of reading the numbers, but they’re available, completely legible, and of course alleviate any confusion to the underlying values. It’s unfair to discredit the chart asking us to not “cheat” by using a feature that it actively and effectively employs, particularly when you do the same on the bar chart.

Also, something I just noticed, while you assert that we’re not perceptually good at comparing areas, we may be better at aggregating them. For example, is Bailing a larger chunk of the budget than Spending, or Giving more than Earning? I don’t think I can clearly answer that from either chart, but I think the area chart gives a slightly better chance. (This idea alone has had me staring at the charts for some minutes — I’d be interested in any real tests on that).

Not trying to be overly critical or negative — I was mostly just taking the challenge you proposed in the last paragraph. I completely agree that the dark budget box is out of place, and the bar chart is far better at showing how it aligns. I also agree that the treemap’s random arrangement of boxes is distracting, and the nicely sorted bar lengths are visually appealing and certainly make comparing two bars trivial.

Still, it’s a tough sell, to me, that the bars are a complete win, and particularly compared to some charts such as the ridiculous GE ones you pointed out last week, I think the Guardian’s treemap, while flawed, is reasonably good.

I wonder how the information would be received had you designed a bar chart showing the differences in pounds between the U.K. budget deficit and each of the other “Big Amounts of Money”. I wonder if it would have communicated the information equally or more effectively then your bar chart?

Or perhaps instead as the difference in pounds it could have been expressed as a percentage difference from the U.K budget deficit. Expressed this way you wouldn’t need to indicate that the values were in (Billions of Pounds)

Anyway food for thought…

Stephen, perhaps the tests you propose aren’t adequate or maybe incomplete, in the sense that it could exist some tasks that are easier to perform with the 2D display. I recognize at least one thing that’s very easy to do in the McCandless surface and very hard in the bar graph: to visually ‘select’ one rectangle and then look to others and compare surfaces many times (back and forth). Try this: focus on Defense Budget and compare it with 10 other values no matter where they are placed. It’s always easy to return to the departure point. If you try the same task with the bar graph you’ll see it’s very likely to get lost, and the effort to return to the starting point is hard. Individually each comparison may be less accurate in the surface map, but you can perform much more comparisons.

In general terms I believe the treemap (actually is just a quadrification) enables very diverse comparatives. The eye goes around selecting different pairs of surfaces or even blocks (some are clearly suggested by the author by color or composition means). The bar graph forces a linear lecture (thus proposing only consecutive comparatives).

Pleasure lies on this non-linearity reading or ‘perceptual freedom’ and I believe is part of the success of this graph. And by success I not only mean the fact that a lot of people shown interest on it, but also that a lot of people read it and had an insightful experience.

Chip,

Thanks so much for the thoughtful response. I appreciate it when people challenge what I say in substantial and reasonable ways, which describes your comments well. I’ll try to address your points one at a time.

It is true that this particular example of McCandless’ work is better than most, but only because little is required of a chart to make his point, which is that the budget deficit is really big. His chart manages to say this, but he could have said this better. I think the bar should be set higher (pun intended).

You pose the challenge: “While it may be true that comparing areas for nearly-similar values is difficult, it’s made easier by this large coverage area.” Actually, your claim isn’t accurate. The fact that these rectangles are relatively large makes them easier to see but not easier to compare. Also, it is not only true that comparing nearly similar areas is difficult; comparisons of all areas is difficult. For instance, a rectangle that is precisely half the size of another rectangle will not appear half the size to our senses if it they differ from one another in both height and width (that is, in two dimensions). A bar that is half the size of another (varying in one dimension only), however, we can perceive to a fair degree of precision.

Your next claim regarding size, however, is spot on. You correctly point out that larger areas of color are easier to discriminate than smaller areas. Using the full-sized version of the bar graph, however, the bars are not so small that the colors cannot be discriminated. Neither chart works well in the small version. In fact, the small version of McCandless’ chart is completely unreadable.

Furthermore on the topic of color you say: “The larger color coverage allows me to more quickly search for colors. I can identify very quickly which boxes are ‘Fighting’ and which are ‘Giving’, but a scan of the bar chart does not make that as easy.” It is true that larger areas of color are easier to spot than smaller areas, all else being equal. Neither chart is designed to help us quickly find items in a particular category. The assumption is that this isn’t useful, given the purpose of the chart. If it is useful, then rather than trying to get the main chart to support this purpose, which it could never do effectively, a second chart specifically designed to support this use would work best. A common mistake that people make when telling stories with charts is to attempt telling the entire story with a single chart. Finding items in the same category would be supported well with a bar graph that groups bars of the same category together.

I didn’t include nested values, such as the bailouts of the two banks that account for 56 and 48 billion pounds respectively of the 289 billion pound total, only because this didn’t seem to add anything useful to the story. I could have easily shown these parts of the total bailout of banks by including them as segments of the bank bailout bar (i.e., as a stacked bar). This would enabled comparisons to the deficit more effectively than the rectangles in McCandless’ chart.

Here’s your next point: “Back to the area vs. length argument, you disapprove of reading the numbers, but they’re available, completely legible, and of course alleviate any confusion to the underlying values. It’s unfair to discredit the chart asking us to not ‘cheat’ by using a feature that it actively and effectively employs, particularly when you do the same on the bar chart.” I included the numbers on the bar graph only because McCandless included them, based on the assumption that precise values were useful. The difference between McCandless’ chart and mine is that mine would work without the numbers, making comparisons of the values possible. It is fair to point it out when a chart includes numbers as text to make up for the fact that values cannot be effectively compared without them. Having to read numbers to compare the values adds time and effort to the process that shouldn’t be necessary with a graphical quantitative display.

Lastly, you make the point: “While you assert that we’re not perceptually good at comparing areas, we may be better at aggregating them. For example, is Bailing a larger chunk of the budget than Spending, or Giving more than Earning? I don’t think I can clearly answer that from either chart, but I think the area chart gives a slightly better chance.” What you have discovered on your own is the one thing that research has demonstrated as an advantage of a pie chart over a bar chart for part-to-whole displays. It is easier to compare the sum of a set of slices to the sum of another set of slices in a pie chart than it is to compare the summed lengths of bars in a bar graph. While this is true of pie charts, as one form of area comparison, summing the sizes of the rectangles in McCandless’ chart is more difficult. If the sums of things ought to be compared, the best way to present them is to pre-aggregate them and then show the results in a bar graph. For example, if it would be useful for people to compare the sum of these amounts per category (bailing, spending, etc.), the following graph would support this effectively.

Something that I didn’t mention in my critique of McCandless’ chart is that it suggests something about the data that isn’t true. Treemaps, like pie charts, are part-to-whole displays. When I first looked at The Billion Pound-O-Gram on McCandless’ website, I assumed for quite a while that the values were parts of some whole, such as total UK government spending. The reason is because I assumed he used a treemap properly. People would be less likely to make this mistaken assumption when looking at a bar graph.

I have to confess that when I looked at the McCandless chart, my first impression is that the Budget deficit does not look all that big compared to some of the other big boxes on the chart and particularly it is not that big when compared to all the other boxes together, so maybe the Budget Deficit is not such a problem after all. I have not read the article so I do not know if that is the impression that it is supposed to convey, however I suspect that it is not.

My second thought arrives a few seconds later, what do these numbers mean? and what does it mean to compare them? The chart compares assets with spending and then mixes up years to further confuse matters. There is Mortgage lending for 2007 at the height of the boom and mortgage lending in 2008 at a much reduced level, but why are they on the same chart? Then there is Trident which is spending over a 25 or 30 year period, which does not compute when compared with spending in a year.

A treemap is used to compare parts to a whole when there are many parts. Putting these unrelated numbers into a treemap like structure suggests that they are parts of a whole which they are certainly not.

I’m a long time Few fan, but also a long time McCandless fan, and I think I’ve finally figured out why that’s not an oxymoron.

The bar graph does a great job of allowing a motivated individual to compare things quickly and deeply. But it looks like every other bar graph in the world. There’s no context integrated into it. So it fails as an infographic.

McCandless’s area chart has a visceral quality that is totally lacking in the bar graph. Those skinny lines don’t carry much weight. They don’t feel big. I look at it and I think: that bailout is *big*. I imagine that chart painted on the wall of an art museum, and people walking away with an emotional sense of the heft of the bailout. If I wanted to give someone a sense of how big the bailout is, I’d use the area chart.

The bar chart provides no sense of proportion. That is, they have no gravitas. They’re the same type of bars used to describe the market share of fava bean suppliers in Iceland. In contrast, big, bold squares and the way they fill the page suggest that the quantities involved are huge. People have a hard time telling the difference between millions and billions. This type of chart does it.

Point being, only the area chart is a good infographic. It draws otherwise unmotivated people into the graph, and it provides context through its visual heft. The line chart, while superior in all the ways Few describes, fails as an infographic because it provides no such context.

What is the goal of the original diagram? Is it to engage the viewer into exploring the information or is it to compare minutia. If the goal is to compare similar values then it fails but if the goal engage the most people into exploring the information it succeeds wildly.

The original diagram is like story that invites people in for fun ride and educational ride where as the simple bar chart is not as visual compelling and reaches fewer people.

@Chip: kudos for trying to break down why the original might be more visually compelling. It’s much harder to analyze why something is visually compelling then to check if it’s measurements are correct and if it follows well known data display conventions.

Stephen,

Your last comment is far and away the most useful part of this conversation. Your bar chart informs, but is not interesting. McCandless’ chart enlightens, but is not strictly correct.

But again, your comment regarding “entertainment vs. information” is more interesting. Case in point, right now I have:
– 3 computer screens in front of me, one of which is constantly updating with live financial data (US Markets seem stronger than they should be – bulls are out in force today.)
– A television on mute tuned to woman’s world cup soccer. (France is up 1-0 on Canada, but I’m looking forward to the US vs. Colombia on Saturday.)
– A radio behind me tuned to NPR (Greece’s austerity measures would piss me off too, but I just don’t see an alternative – and the recent vote gave a hopeful bump to the EURO.)

This is what you are competing with. Your bar chart, while a better way to encode this information, absolutely will not give me pause in this environment. If you want to tell me a story, you best find a way to convince me to give you some attention first. It’s not about entertainment, it’s about breaking through the noise we’re all working with today.

I’m not going to argue in favor of the way McCandless encoded the information – as you and others have pointed out, his chart just isn’t technically very good. However, something made me take the time to look at it. I looked at it the first time I saw it, and it was the reason I looked at your post here. And both times I walked away with a bit of new knowledge.

The New York Times of course does this better. They capture the imagination, give me pause, AND deeply inform. Clearly that is the target. But I think you miss the deeper issue if you propose that your bar chart is the answer.

Stephen,

I appreciate the response.

However I think I need to clarify.

First: enlighten vs. inform. Indeed, to enlighten one must inform. McCandless’ chart does enough informing to enlighten me on a few points. I suspect that’s all he was looking to do. Could he do more? Should he do more? Absolutely. I agree with everything you say on that. But by stripping away anything attention-grabbing, your bar chart sadly does neither.

Second: The bar chart neither informs nor enlightens because I wouldn’t stop to look at it. I agree 100% that “the solution to overwhelming chatter is not to reduce the quality of data visualization..”. Again, I’m not defending the poor sensemaking quality of McCandless’ graphic. However you say “the primary goal of data visualization should not be to cut through the noise…” and this just strikes me as naive. Perhaps this is the case in much of the work you do with dashboard design wherein the user is expected to provide you with undivided attention. But as you yourself point out, McCandless is operating in journalism space – where he better cut through the noise lest he be tossed into space with all the other ineffective information sources. In this sense, his graphic succeeds here where a bar chart would not. That alone is my core point.

Finally regarding the NYT: I didn’t say they produce infographics primarily to entertain. Indeed, they produce to inform – but they seek, and find, the balance. They recognize that to gain readership they need to capture attention. They realize they are competing with the likes of this:
(picture taken from my chair seconds ago).
And their work clearly and regularly demonstrates this simple fact.

DR,

Your response confirms my concern. We appear to be living in a time when you and many others will not attend to an information display unless it exhibits eye-catching visual effects, even if those effects diminish clarity, accuracy, and breadth of understanding. You claim that you will not even look at a chart that displays information simply (which in part means no unnecessary visual effects), clearly, accurately, and in a way that leads to greater comprehension.

My insistence that the primary role of data visualization is not to cut through the chatter and grab your attention but to inform is not a sign of naivete but of principle. I believe in the potential of data visualization to enlighten far beyond the low-density and often inaccurate messages that McCandless’ charts achieve. Designing data visualizations to feed your demand that something scream loud enough to be heard above the chatter is a practice that undermines the power and goal of data visualization. As a practitioner who has dedicated his professional life to data communication and sense-making, I have no choice but to oppose this current trend.

If we redefine the primary goal of data visualization as attention-getting to cut through overwhelming chatter, the chatter will only increase. As it does, what will we require next of data visualizations to cut through the even greater chatter? Perhaps McCandless’ heir will develop a technique for shooting bolts of electricity from charts directly into the hearts of readers to stun them into attention. People need to be informed. The answer to distracting chatter isn’t to scream louder and louder. The solution to a world filled with ever-increasing noise is to reduce the noise.

I’m noticing a trend among supporters of McCandless-esque infographics. Many apparently think that data visualizations need to “grab the attention” of people who otherwise don’t care, and this property seems to be the only basis for preferring McCandless’ impoverished treemap over a simple bar chart.

I can’t seem to figure out why engaging apathetic passers-by is a higher priority than informing people who already do care, are already interested, and will already be engaged when the data are presented such that the viewer can make sense of them. When you put McCandless’ “art” in front of them, it’s like a slap in the face. So much information is hidden from the viewer because the data are presented with inappropriate display media; all the same numbers are there, but data visualizations can (should) encode a lot more information than numbers (easy example: rank, which the viewer must search for and discover on their own in McCandless’ version). When such useful information is hidden, the viewers are left with only clues, and must recreate it themselves.

If the priority is grabbing attention, there are much better ways than obfuscating the data with “art”, as McCandless work seems to do.

Assuming that we are perhaps nearing the end of this particular discussion, let me quote the words of a man who shares my concerns:

If an editor should print bad English he would lose his position. Many editors are using and printing bad methods of graphic presentation, but they hold their jobs just the same. The trouble at present is that there are no standards by which graphic presentations can be prepared in accordance with definite rules so that their interpretation by the reader may be both rapid and accurate. It is certain that there will evolve for methods of graphic presentation a few useful and definite rules which will correspond with the rules of grammar for the spoken and written language. The rules of grammar for the English language are numerous as well as complex, and there are about as many exceptions as there are rules. Yet we all try to follow the rules in spite of their intricacies. The principles for a grammar of graphic presentation are so simple that a remarkably small number of rules would be sufficient to give a universal language.

These words are from the book Graphic Methods For Presenting Facts by Willard C. Brinton. He wrote them in the year 1914. At the time, it was true that “useful and definite rules” for graphical presentation had not yet evolved. Today, this is no longer true. They will certainly evolve further, but not very quickly if we continue to repeat the mistakes of the past.

It is this sense of history that adds fuel to my passion. We can do so much better. We should do better, but we backslide every time we become enamored of some new technology or some new group that involves itself with data visualization without learning the lessons of the past. How many steps backward must we take before we regain forward momentum?

DR,

Journalistic infographics should support the telling of a story. They need not and usually cannot tell the entire story by themselves. In the case of journalism, telling a story is what must be done to inform. My problem with McCandless’ infographics is that they don’t support the telling of a story well.

The title of this blog piece is “The Billion Pound-O-Gram Redesigned.” I understood “The Billion Pounnd-O-Gram” to be McCandless’ description of the chart. My intention was to reproduce his chart, not the entire story. I’ve described in broad strokes what I would do if I retold the entire story, but won’t bother actually doing this, because time is short and I don’t find the story itself interesting enough to bother.

I must admit that the design of the McCandless graphic initially grabbed my attention. I studied the chart for a while before I realized that it wasn’t showing parts of a whole and that items like “Africa’s entire debt to Western nations” were completely arbitrary– McCandless could have just as easily used the value of all the oil Saudi Arabia.

That realization led to an overwhelming feeling of being duped, and had I stumbled upon this graphic somewhere besides this forum I would have been pissed off at McCandless for wasting my time.

Some of the supports of McCandless seem to feel that the goal of infographics is to provoke an emotional response in viewer, not to inform or encourage intellectual exploration discourse. It appears to me that they are confusing art and journalism.

>
If I were an editor I would struggle to choose between those two options. My thinking would be:
-” The McCandless option looks good, and my readers will stop and look at this. But ugh, it doesn’t actually make sense.”
-” The Few option has the right information, but ugh, the ad printed to the left will garner more attention.”
>

DR, are you an editor for UK’s “The Sun”? Is there a need for a plunging neckline in an article about how a billion pounds are spent?

As for concerns about an ad distracting attention away from Few’s version, I like one reader’s comment posted in reference to the online version of the Guardian’s article:
“I can’t read the f[****] graphic because of a Lowe’s ad. Wonderful”

Good call on the areas, but I do believe it’s a simpler explanation: It just looks good (to McCandless). As a programmer, he should have the mathematical skills to calculate the areas correctly; it’s not that difficult. Since he fails to plot correctly even in one dimension, I suspect it’s because it looks good. An example: See http://www.informationisbeautiful.net/play/snake-oil-supplements/ where the Strength of evidence seems to be a continuous variable; by eyeballing it, I can count at least 30 different values on the y-axis. However, checking the data, there are only 10 discrete levels (0-6, some in half-steps). There is no way that erroneous plotting due to a bad algorithm.

Many things that pass for infographics these days uses ineffective graphing techniques are naive and dumb down complex issues in an attempt to engage the ignorant. Much so with McCandless, for example the Billion Pound-O-Gram is not an effective graph, and it lacks the most important variable to make sense of the budget deficit; the total size of the budget, as pointed out on this blog. The world is multivariate, so comparing data and putting them in a meaningful perspective is the most basic task in sense-making. McCandless adds a “pseudo-multivariate” feel to it to counter his dumbed down selection of data by using different shapes and colours (the shades of green in the Billion Pound-O-Gram does not code any information).

However, the typical McCandless infographics have some other features that are somewhat more unique to him: The total disrespect for or lack of understanding of the data. The “Planes or Vulcano” (http://www.informationisbeautiful.net/2010/planes-or-volcano/) graphic is not calculated correctly (it’s a snapshot of one day, almost worse case scenario for the cancelled planes and not the complete emission from the volcano), does not it include an estimate of how much extra CO2 emitted due to increase of alternative transportation or how much of the travels simply where postpone. The worst example I’ve found so far, because it’s closer to me, is the Snake Oil graphic. It’s supposed to be a summary of the scientific evidence for health supplements, and he even puts in a “worth it” line. But the general consensus among scientists is that health people with a recommended/ normal diet do not need any supplements. So where does he get that line? A quick look at the data reveals that McCandless does not know the different between lab bench experiments, supplements given to counter a illness/ deficiency and healthy people taking supplements in combination with a normal diet.

In short: Even if the reader can see past the ineffective graph, naive comparisons and wrong areas/placement along axis and manage to extract some information to take away, chances are that this information in many cases is wrong. That’s why a redesign as shown in the blog-post above does not make a good infographic; it does highlight the ineffective charting technique, but don’t deal with the more serious, underlying flaws.