Visualization is not a very clearly defined field. There are many variations, ways of doing it, and ideas around it. That is valuable, because it keeps the field moving and brings in fresh ideas. But it also brings with it people who like using visualization’s tools and talk about visualization, but what they are doing is something else. We need to start calling these things what they are: a cargo cult of visualization.
John Grimwade describes how his students are getting confused: My students keep bringing stunning examples they have found on the Internet. But they are rarely able to tell me what the graphic shows.
Say what you want about visualization, but its central idea is clearly the representation of data. Not being able to recognize what the data even is, and only being drawn into the image because it looks cool is a problem.
The term cargo cult describes a kind of religion or cult that is based on imitating behavior with the goal of material gain (the “cargo”). It started with relatively primitive societies observing the behavior of more technologically advanced ones, in particular during World War II on Pacific islands. After American or Japanese airbases had brought in fascinating new technology and material wealth (relative to what they had seen before), some cultures started building makeshift airports with towers and radios from wood, and mimicked the behavior of the airport personnel they had seen. The idea was that the behavior was some kind of cult that attracted the cargo, so imitating it would bring it back.
Richard Feynman coined the term cargo cult science, by which he meant work that is using scientific methods but that still fails to produce science. This even includes scientists, when they are not honest about their biases and let them taint their work: just going through the motions without understanding the actual issues is not enough.
Today, the term is used broadly for misappropriation and use of techniques without understanding what they are for. Just because they were successfully applied in one area does not mean they will also work in another (e.g., periodic systems).
Common chart types are so widely understood that they are now used for jokes. The website GraphJam is a prime example, there are flickr groups that “visualize” song lyrics, FlowingData’s Data Underload, etc.
One of my students also has this rather clever Van Diagram t-shirt.
There’s nothing wrong with these, but they need to be understood as using the tools of a field for something that is not its goal. That may be obvious when the topic is van diagrams, but in other cases it’s a bit less clear.
Mountain Top Stock Charts
Perhaps the most obvious recent example are Michael Najjar’s mountain top stock market charts (image at the top of this article). Najjar uses photographs of mountain tops and shapes them so that they represent stock market data. He argues that the virtual data mountains of the stock market charts are sublimated in the materiality of the Argentinean mountainscape.
This is, quite simply, nonsense. There is a place for art, and there is a place for visualization. Mixing the two is difficult and dangerous, and often leads to things that are neither. Throwing around art jargon doesn’t make this any better.
The discussion about the sublime and how it might apply to visualization is outside the scope of this little article. I wrote a paper discussing this issue a while ago in case you are interested, and I am planning a posting or two about this topic here.
But the point is that the mountaintop images entirely obscure the data. This is not visualization. It may be art, but it’s not some kind of hybrid of visualization and art. Just because data was involved at some point does not make this a visualization. The problem here is not that this was done, but that it claims to be something that it is not (unlike the poetry “visualization”, for example).
Not to be mistaken with FlowingData, these two books collect infographics and visualizations. They are presented simply as images, without context such as what the data or task even were.
It’s telling that even Andrew Vande Moere, who does not tend to be overly critical, felt that the case for the insightfulness of the presented work was overstated.
This is the type of presentation you can also find a lot on the web though, where people post pretty pictures from visualizations without any interest in the data or the actual expressiveness of the work. And unsurprisingly, most of them are pure junk.
The Ford Fiasco
Ford went all out to produce some pretty visuals for their Ford Fusion. They organized a kind of contest where they had six teams compete against each other. The tasks are mostly nonsensical, and seem to have been picked simply to generate data to transform into some of the worst three-dimensional bar charts I have ever seen (plus, of course, gauges and a kind of heatmap).
Visualization has become the tool to create pretty (or at least colorful) visuals that attract attention. There’s about as much visualization in these images as in an actor depicting a scientist in a tv commercial.
Why We Need to Fight This
While the visualization community is still largely refusing to draw any lines, we are letting others blur the distinctions. To many people, visualization already is primarily about being pretty and colorful, and the data representation is only an afterthought. At the same time, people like the Data Flow editors are talking about insights when they are not even providing any kind of context for the images in their books.
Visualization needs to be more clearly defined, not less. It needs more limits, not more sprawling inclusion of all and everything. We need to start drawing lines in the sand or it will be too late.
25 responses to “The Visualization Cargo Cult”
Excellent post, and it reminds me of the Information Visualization Manifesto by Manuel Lima: http://www.visualcomplexity.com/vc/blog/?p=644 and Data Visualization: Keeping the Story Straight by Stephen Few: http://www.perceptualedge.com/blog/?p=613
Thanks for a great critique of this problem. I just finished a training presentation that made the exact same point. I think we need to be more forceful in our assertions about what are good and bad data visualizations — at least in terms of their ability to effectively communicate data.
And I appreciate adding the term “cargo cult science” to my vocabulary. The place where I part ways is at the “we have to draw lines before it is too late” part. I am a little unclear on the risks, or what will happen when it is “too late.” It seems that it would be very valuable to get much more specific on this point.
The trouble is, if you focus on drawing lines in the sand and organizing everyone into two camps, you end with a war over sand lines. This can be very wasteful and distracting. I do think there is a lot of value in distinguishing between different types of visualizations, and making it clear that there is a difference between graphics that inform and nonsense pictures. Lines are good, but I prefer more of a loose taxonomy. Insisting on hard boundaries between “real” and “not real” visualizations seems like it is just seems to me to be inviting unnecessary trouble.
Very well put – that’s pretty much exactly how I feel.
@Yasmin: but is it more wasteful than losing potential interested people that look at the clearly bullshit graphics and think “There’s no way these visualization guys are serious.”?
People that using InfoVis-like Tools create something that doesn’t describe the data itself but is used to make “art”. Are they actually creating InfoVis stuff?
Disclaimer: this comment is very pedantic.
Great post, and I’ve seen a few similar ones (eg Stephen Few a while ago).
However, I have an issue with your definition. Visualisation is visualisation. Full stop. If someone takes data and makes art out of it, they have visualised it. You and I (and probably the artist) would agree that they may not have turned it into information, but the term Visualisation (or “data visualisation”) cannot and should not be mandated to cover that too.
So, what’s the relevant term for visualisation that informs or tells a story? Well, I don’t know the answer to that!
“Informative visualisation”? Ugh – that’s horrible.
“Visual analytics” – maybe Tableau are on the right track?
Great blog, by the way!
Follow me on twitter!
I think they key thing to understand is that Information Visualization is a medium, just like any other. As with any medium, people will use it to present results, to influence others, to disinform, for artistic purposes or to produce utter crap. There is no way the ‘scientific’ community will be able to police all content that is self-labeled ‘information visualization’.
I see the rise of ‘nonsense-graphics’ as a direct result from the mainstreaming of the infographics medium, which is a good thing in general. To give an analogy: I can point to some of the worst sites on the web in terms of content, design and markup. Some of these don’t even carry any “informative content”! Rediculous! Down with them!
Visualization needs to be more clearly defined, not less. …c’mon…that’s like saying we need to define music, or writing.
We need to start drawing lines in the sand or it will be too late. Too late for what? To define it? So what?
Great post on a topic I’ve been pondering as well.
There is definitely somewhat of an interesting relationship between art and visualization. While the best visualization for data consumption may be a Tufte-esque simple graphic, that sort of graphic would rarely get passed along or shared enough to reach a significant audience, as some of these other “art pieces” do regardless of how well they represent data.
Adding some artistic elements to a visualization is — in a sense — a way to market the data. It can make the data more fun to consume, digest, and discuss. Where I draw “the line” is when a visualization is created for art’s sake, and the data is in the passenger seat, as you mentioned. Art should be added to data, not the other way around.
I see David McCandless’s “When Sea Levels Attack” as a good combination. The visualization tells a story, has great viral reach and sheds light on a real-life issue.
I’d be interested in seeing what visualizations you and others feel approach the line in the sand, but stay on the side of representing data.
See http://www.flickr.com/photos/philgyford/4505748943/sizes/o/ for an amusing take on the glut of “infographics” – I think in direct relation to the iPad unboxing graphic that’s been circulating.
I agree with Frank. I don’t like the idea of policing visualization, in fact, it’s likely impossible and would be a waste of time. Just as with any other medium, there are good examples of it and bad examples. There’s good journalism and bad, good and bad art, film, websites, ad infinitum.
Isn’t our time better spent creating good visualizations, teaching others about it, raising awareness about poor visualizations, and so on?
Also, I want to argue Michael Najjar’s work. And to be fair, he used the word “resublimate” not “sublimate.” I do feel his works are visualizations, no matter how loosely he interpretted the data. I really enjoyed his High Altitude series. If anything, data art makes me think *more* about data and the possibilities of what it’s telling us. Perhaps more metaphysical than factual, but I think that’s needed in the world too. For me, his artwork conjoins the awe-inspiring nature of spectacles of nature and high finance — both larger than life. I think it’s an interesting juxtaposition of emotion and fabrication. I don’t think he ever said, “use this manipulated photograph as a chart to gain insight into investing.” He makes no conclusions from the data and presents his artwork merely as that, artwork.
I think the best way to define the field is to have a community that values and consistently produces good, solid, perfectly “boring” visualizations. It seems to me that the academically-minded people interested in visualization spend more time designing, evaluating, etc. the techniques to be used in visualization than actually producing data visualizations. On the other hand, most of the “visualizations” that get attention are meant to be artistic, which might give the impression that most people that produce visualizations “in the real world” ignore the concrete goal of informing the reader above all other concerns.
To give a simple example, every single visualization I have ever seen from the NYT is great. *These* guys are doing visualization. At the same time, every single picture from Data Flow is *not* visualization, as far as I’m concerned. I’m not claiming there isn’t a public that derives enjoyment from the Data Flow examples. I *am*, however, claiming that there isn’t a public that derives insight from it.
There is no problem with saying that art is inspiring, or that it makes a real point about the real world. Still, no one would claim Picasso or Nick Ut are political essayists. The moment an artifact becomes closer to *Guernica* than to *Leviathan*, it simply ceases to be a visualization, and becomes art.
(Sorry for spamming the comments, robert :) How about this for a rough definition?
If it elicits emotion, then it’s art. If it elicits reflection, then it’s visualization.
Thanks for this enthusiastic article. I agree with you concerns about the quality of visualization and the readability of the information that lies within. I take your thoughts as an advice to ensure enough perspective and context of the represented data in the visualizations I present on Datavisualization.ch.
One more thing comes to my mind in regard of selling the data: The NYT team achieves engaging visualizations with their ability to make storytelling a main objective. Rather than prettify the visualizations with decorations of any kind, they introduce established journalism best practices like guiding the reader along the data or using related media for imrpoved understanding of the facts. So, storytelling with data seems for me the right route to engage a broad audience.
I appreciate your thoughts into this and agree that there are a lot of data-like visualizations that can make us frustrated… so is the nature of anything when it hits a mainstream (i.e. music, painting, etc..).
However, I really feel that some of this is just too one-side and that you are falling victim to your own frustrations. We both agree that we have to push forward and keep exploring… but why are the mountain top stock charts utter nonsense? I don’t agree with this notion for a few reasons.
For decades now, we are discovering more and more a relationship between many seemingly disjointed things in this world. How, for example, froth in a cappuccino can mathematically represent a pattern similar to a galaxy.
Stock markets have been aggressively analyzed for a good century now. So much that we owe much of the advancements in our technology to companies, such as IBM who have focused much of their attention on market behavior. But we are still far from understanding it. There are many many models to try to define, predict and explain market behavior, but every model has it’s flaw. Yet somehow, you can create fractal stock market patterns (http://classes.yale.edu/fractals/) that, though fake, strongly resemble a real stock market pattern… and somehow the mountain top stock photos strongly represent this too… is this absurd? I don’t think so. It’s only as absurd as pre-20th century math that had a very limited understanding of patterns in nature.
As data visualization experts in the field, we should be able to separate the cute, meaningless infographics from the purposeful ones. The line between the two is blurry, but I don’t think we should be saying we need to lay down more rules and disclaim what seems absurd because it doesn’t follow rules we understand.
The idea of cargo cult is a good one. There’s plenty of that… but I wouldn’t get a cargo cult mixed with someone exploring “exotic” notions. Before we launch a probe to mars, we spend countless hours building virtual scenarios to better understand what challenges may lay ahead.
Yes, fake, shallow visualizations can frustrate. But the more I see of them, the more I know what to avoid…
Firstly, it is good to see such a set of reflections on this post. I believe that there is space for very functional oriented visualisations and purely artistic visualisations under the banner of data visualisations. Most peoples/studios work encompasses pieces that fall between the two ends of a spectrum and both sides, art and functionality, inform the other, resulting in, hopefully, explorations that move the area forward.
Data Flow 1, I haven’t read DF2, does indeed have much Cargo Cult Viz work and at times the magnesium like flare of interest the internet showers on visually engaging but uninformative pieces can frustrate. This is however, the pot calling the kettle black as interest in our work as been for both data driven more artistic visualisations and functional visualisations.
I don’t think that drawing a line, in what I see as a spectrum, is necessary. Artistic oriented data visualisers such as Jer Thorpe are producing, as expected, pieces with more focus on art vs being informative but at the same time, they produce tutorials and works that help people move into the field and spark the interest in data viz which can only help the industry as a whole.
Clients, customers, and audiences deserve to use interactive data visualisations that are both beautiful and informative and exploring the functional/artistic spectrum can only be a positive thing.
The challenge, as practitioners, is to communicate to our customers the area of the spectrum in which we are strongest and to allow artistic data visualisers the opportunity to create and market their pieces and for the more functional visualisers to receive recognition, business etc for their pieces that may not be as shiny.
1. In this time and age, visualization is a medium. Like music, there exists different tastes and directions, mostly happily together. The vast majority of people, government and businesses are able to make the distinction between artistic and scientific approaches quite accurately. Convincing readers that people misbelief the neon Ford graphs and stock market mountain tops to be Beethoven instead of Lady Gaga and Sting is just not a very believable argument. Let alone that none of the examples you mention even attempt to claim they aim for delivering insight. But visualizations they are nonetheless.
2. While your argumentation deals with the typical tension between art and science, it is still surprising to me you show no appreciation or respect of what art and culture actually constitutes. And this for someone who strongly argues people need to understand the actual issues before getting involved. Associating words like “nonsense”, “throwing art jargon”, “pure junk” with works that have potential artistic value (and yes, “value” is used appropriately here) is not a particular constructive way of starting any sort of dialogue with those people on the other side of your “lines in the sand”. But something tells me that finding any sort of common ground is not your goal at all.
Great post and I’m completely with you regarding the stock “charts”. The fact that it’s beautifully done doesn’t make it any better or more informative than this http://www.cc.gatech.edu/gvu/ii/infoart/ (obviously I’m not a “fan” of this kind of “Ambient Information Art”).
However, I don’t think we need to draw any new lines in the sand. IMHO Information Visualization is a pretty well defined term and field. This allows us to tell if something a) is infovis at all and if yes b) whether it works well or is poorly done (and tell why).
For me simply using the term visualization seems far too general — visualization on its own could even be an image you make up in your mind only.
Data Flow 2 is an inspiring and captivating book, which despite a lack of deeper analysis provides a stimulating showcase of ideas. The most puzzling aspect of the book is the apparent disconnect between the foreword and its remaining pages. As Andrew Vande Moere stated in his review, the book is simply not the best evidence for the claims of relevance and insightfulness portrayed in the foreword. And this apparent confusion, which might seem odd to some readers, underlies the students’ puzzlement that Robert has exposed in his post. A manifestation of the same disorder can be found here:
But any attempt at resolving this confusion has to come through criticism and exchange of views. The argument of visualization as a medium is a potentially interesting one, as long as it invites discussion and doesn’t work as a defense mechanism for debate. After all, experimentation is as important as criticism. How else can we evolve and mature as a discipline? And because criticism is important, here are a few more posts to add to the discussion:
… the notion of a cargo cult is _exactly_ about confusing “exotic” notions with bullshit, of which your comment seems to be guilty: “It’s only as absurd as pre-20th century math that had a very limited understanding of patterns in nature.”
There are reasons for which people are interested in fractal patterns in stock markets. It has to do (for example) with scale-invariant properties, and the mathematical predictions that can be made from it. It has got nothing to do with the fact that stock markets look like mountain tops or snow flakes.
*The vast majority of people, government and businesses are able to make the distinction between artistic and scientific approaches quite accurately.*
*While your argumentation deals with the typical tension between art and science, it is still surprising to me you show no appreciation or respect of what art and culture actually constitutes.*
Strawman. The issue, fundamentally, is that the visualization cargo cult artifacts confuse art and science, which is exactly what happens when there is a claim that a picture of a mountain can bring real insight about stock markets. The post says: “But the point is that the mountaintop images entirely obscure the data. This is not visualization. It may be art, but it’s not some kind of hybrid of visualization and art.”
Also, notice that “pure junk”, in the post, did not even refer to the Data Flow books. But let me state as clearly as I can: as far as visualization value goes, the Data Flow books contain mostly junk. As far as art goes, they are *art*. There you go, now you can be angry at me instead.
Visualization just means a visual representation of data. There is no requirement that one visualize important data or elucidate an important fact through visualization. You can attack the Ford ad for visualizing unimportant data, but then you miss the point of the ad: namely, that it’s an ad and not a scholarly product. The stock-market mountains are a perfect sublimation of the raw data that generates them: they transform a set of data that is mostly misunderstood by viewers into something they can gaze at. The sublimation should _require_ that the information that could have been imparted be discarded.
This is a very silly discussion. Unless a visualizer produces the data he visualizes, he is merely an illustrator. Whether the illustrator illustrates useful information or junk doesn’t really matter. He’s just an illustrator, not a statistician. He’s an aesthete, not a scientist. He’s not terribly important to the data.
I think it is well defined…I mean visualization field..there is data and there are graphics…why are we over complicating and applying a mantra of art mystery to something which is straightforward?! Thought provoking piece by Robert though