Virtual Globes are a seriously bad idea for thematic mapping

by Mark Harrower on April 17, 2009

Google Earth is amazing. As we’ve commented here before, it continues to blow our minds and has also done wonders for the popularity of maps. And let’s be honest, it looks super cool. There is no doubt that Google Earth is much sexier than that boring old atlas collecting dust on your shelf: It’s interactive, seamlessly integrates distributed data sources, animates the surface of the earth over time, facilities virtual communities, can be customized by both developer and user, etc, etc. It’s hard to not be impressed.

So all of our maps should be in Google Earth, right?

Wrong.

In fact, despite recent efforts to create a suite of thematic mapping approaches, Google Earth is a terrible environment for presenting many kinds of thematic maps. I’d go so far as to say that the 3D prism maps and 3D graduated symbol maps we see popping up in Google Earth are pure chart junk, of the kind Tufte warned us about repeated for past 25 years.

3D prism map of population in Google Earth

3D prism map in Google Earth (blog.thematicmapping.org)

3D human figures as proportional symbols

3D human figures as proportional symbols (blog.thematicmapping.org)

CHART JUNK

Chart junk takes what should have been a simple-to-read graphic and makes extracting information (1) slower, (2) more difficult, and (3) more prone to reading errors, because of excessive ornamentation and unnecessary design additions—like adding a 3D effect that communicates nothing in and of itself but simply “looks cool.” This is not idle speculation: Research consistently shows chart junk and “redundant ink” hurt otherwise fine graphics.

Want to see for yourself? Download these two example KML/KMZ files from blog.thematicmapping.org and run them in Google Earth. While you’re looking at them try to extract numbers or compare places: KMZ File 1 |  KML File 2

“BUT THEY LOOK COOL”: A TECHNOLOGY IN SEARCH OF A PROBLEM

As Abraham Maslow said, “If the only tool you have is a hammer, you will see every problem as a nail.” This seems to be the case with virtual globes and the developers who love them and insist that any and all kinds of thematic data belong there. Instead, I’d challenge us to take a step back and ask,

WHY DO WE MAKE THEMATIC MAPS?

For a long time folks like Robinson, Dent, and MacEachren have been arguing that thematic maps exist to support two basic tasks: (1) the ability to extract numbers/facts about specific places (e.g., 15C in Paris) and (2) the ability to judge those values in geographic relation to other places (e.g., 5C warmer than London, about the same as Milan). In other words, we want both specific details and overall patterns to be obvious on our thematic maps. And we want all of that AT A GLANCE.

The problem with digital globes (as with all globes) is you can’t see half the planet and, due to curvature, really only about a 1/3 of the planet clearly at once. Which leaves us with a conundrum: If you’re only mapping a small place (e.g., a country), why do you need to have it on a globe? And if you have a global dataset, why would you allow your readers to only ever see ½ the data at once? They can rotate the globe (more on this later) but they’ll never be able to see the entire dataset at once. That makes understanding overall patterns very difficult, and asking folks to “remember” half of a global dataset while they spin the globe to the other side is far, far beyond the meager limits of our working memory. If you’re not convinced, just try it.

KNOW YOUR HISTORY

What makes these recent developments even more frustrating is that in the 70s and 80s, with the advent of digital map making, cartographers flirted with, and largely rejected, faux 3D prism maps and 3D graduated symbol maps (like the two examples above) since they suffered from several limits:

    1. visual occlusion (not all of the map can be seen at once since some places hide others)
    2. people suck at estimating volumes, especially of complex shapes (e.g., try estimating the size of moving van you’ll need for your home)
    3. mental rotation of complex shapes is extremely hard, so hard that it is often used as a measure of intelligence in IQ tests.

Many a thesis and dissertation was written in the past 40 years demonstrating these limits to human visual processing.

The nice thing about Virtual Earths is that you can rotate them, so the problem of visual occlusion is solved, right? Yes and no. Yes, interactivity and the ability to rotate the globe can help reveal hidden places, but no, these virtual globes introduce a significant extraneous cognitive load because the user must now think about controlling the globe (not always easy with a mouse) while also trying to focus on the thematic content. In fact, adding a complex task, like visually acquiring the Google Earth controls and then trying to figure out how to move/scale/reposition the globe between two other tasks effectively “flushes” short-term working memory. It’s a kind of mental sorbet, which is why giving folks something distracting to do is a common trick in memory tests (they lose their train of thought). Why would we deliberately do this to our map-readers?

BIG PROBLEM: INCONSISTENT SCALE

In the examples above it is really hard to judge relative sizes. Why? Because the scale of the symbols is constantly changing, and the ones closer to the viewer are much larger (and at a different scale) than the ones far away. Given that it has been long established in cartography that people are terrible at estimating sizes, and even worse at estimating volumes, it is utterly inane to compound this failure by drawing the symbols at different scales. Of course it is worse than this: Rotating the globe slides each symbol through its own scale transformation path, changing in size with every pixel the maps are moved.

This is an absolute rule: If you want to give people the best chance to judge the relative sizes of objects, they should all be drawn at the same scale.

STILL NOT CONVINCED? LET’S DO SOME USER TESTING

Judging height is critical to the success of this map, yet most heights are obscured

Judging height is critical to the success of this map, yet most heights are obscured

TASK #1: As quickly as you can, how does Nepal compare to Uzbekistan?
TASK #2:
As quickly as you can, find all of the other places on the map similar to Nepal? Which place is most similar? Which one least?

Hard, isn’t it? To be honest, it shouldn’t be: A regular 2D classed choropleth map or proportional symbol map would make short work of those questions. So what did we gain by extruding the countries up into space? Not much that I can see.

    1. The Lack of a zero-line referent makes it hard to judge absolute magnitudes.
    2. The “fish eye lens” effect mean each prism is viewed from a different angle than its neighbors, making comparison just a little bit harder as we have to mental account for these differences in our estimates.
    3. It is hard to judge the height of something when you are staring directly down at it. This matters because height is the visual variable that does the “work” in this graphic—it’s how the data are encoded visually. Why obscure the very thing map-readers need to make sense of the graphic (e.g., the side-view height of each polygon)?

SOLUTIONS?

I need to be convinced of two things: (1)  something is fundamentally wrong with our proven and highly efficient planimetric thematic maps, and (2) that reprojecting this data onto a virtual globe somehow solves those problems. Otherwise, we truly have a cool new technology in search of an application, and that’s just putting the cart before the horse.

Some suggestions: First, unless the 3rd dimension communicates something and isn’t merely redundant data already encoded in the colors, sizes, etc., do not include it (for all the reasons outlined above). Second, if you want folks to perform “analytical map reading tasks” such as estimating relative sizes, distances, or densities, keep scale constant. Third, do not obscure parts of the map behind other parts if that isn’t inherently relevant to the data (e.g., this is fine for terrain visualization). Fourth, and most importantly, do some user testing before presenting a new technique as the best thing ever: It’s how research works and why it is important.

So what things are Google Earth (and other Virtual Globes) good for? The consensus around here is (1) to engender, quite powerfully at times, a qualitative “sense of place” or “immersion”; (2) for virtual tourism (e.g., sit on top of Mt Everest) or virtual architecture/planning; and (3) to perform a kind of viewshed analysis and see what can and cannot be seen from locations (line-of-sight). All of those are inherently 3D-map reading tasks in which the immersive, 3D nature of the map is important. By comparison, population data (one number per country) is NOT inherently 3-dimensional and is only made to suffer when dressed-up in prism maps and 3D figurines.

Cartography, like all good design, is about communicating the maximum amount of information with the least amount of ink (or pixels). The world is just too complex and interesting to be wasting our ink/pixels on non-functioning ornamentation.

There are 15 comments in this article:

  1. April 17, 2009D. P. Huffman says:

    Well put. Feeling devil’s advocate, though, I am prompted to ask what you think about the value of “cool things” in getting people’s attention? If we, for example, make it 40% harder to read the map, but also increase readership by 300%, is it worth it?

    I may be oversimplifying your position, but you seem to say that good design in general boils down ink efficiency. But what is attractive to people is not always the most efficient, and to what extent are we responsible for providing what people want when it is what our journal articles might argue against?

    Of course, the ideal solution is to make something attractive, functional, and ink-efficient all at the same time.

    Just a thought.

  2. April 17, 2009Bjorn Sandvik says:

    Hi Mark,

    Thanks for critiquing my work – this is a timely and enlightening discussion for wannabe map-makers. I also enjoy the jewels in cartographic literature. I’ve posted a quick response on my blog:

    http://blog.thematicmapping.org/2009/04/virtual-globes-are-good-idea-for.html

  3. April 20, 2009rei.lento says:

    Dear Huffman, the main aspect here is the message to be passed to the readers. I believe it’s better to have 10 readers that do get the proper message that the author of the map wanted to pass than having a 1000 readers that don’t get anything. Or worse, that get a wrong message. Although I do believe in the idea of the value of “cool things” but it can’t obscure the message.

  4. April 20, 2009mark says:

    Hi Daniel – This is a very interesting question and I definitely struggle with this balance between absolutely minimal graphics and the desire to make my stuff look nice/cool (I think many designers do)…not that those two goals are mutually exclusive. But, yes, I have added eye candy to many of my maps (like a drop shadow or ink effect to water) to give the graphic a bit of pop, BUT I would remove it in a heartbeat if I felt it was getting in the way of reading the content: The content is king.

  5. April 21, 2009Virtual Globes | PTS Blog says:

    […] 1. Dr. Mark Harrower, Axis Maps, Virtual Globes are a seriously bad idea for thematic mapping. […]

  6. April 21, 2009Tom Auer says:

    Mark,

    Many thanks for taking the time to write this post. It was much needed.

    My biggest problem with thematic mapping in 3D, which you spoke to, is the fact that overall pattern study is demolished. Not being able to interpret the holistic spatial arrangement makes mapping in this manner useless. You can say goodbye to any chance of macro/micro emergence or pattern inference. This extends to network mapping, which has been popular in 3D, because of the ease of drawing curved lines from node to node. However, even in this situation, it’s impossible to navigate the map and remember where lines were coming from or going to.

    Let’s hope that the popularity of the Google Maps API and improved cartographic design for mashups makes their use more ubiquitous and appropriate, eliminating any need to thematically map in Google Earth.

    Regards,
    Tom

  7. April 24, 2009Richard Treves says:

    Interesting post, I agree with most of your points but I think there is a place for thematic maps in virtual globes see my own response post:
    http://googleearthdesign.blogspot.com/2009/04/axis-on-3d-thematic-maps.html

    I’d be interested in tracking down the research you mention in ‘Know your history’, any chance of some references or keywords I can search on to pull up the literature?

    Rich

  8. April 27, 2009Bjorn Sandvik says:

    Hi Mark,

    What is your position regarding this quote from Slocum et al., 2005 (”Thematic Cartography and geographic visualization”), page 70:

    “Use of perspective height produces what is commonly termed a prism map. In Figure 4.11 note that perspective height is the only visual variable receiving a “good” rating for numerical data. The justification is that an unclassed map based on perspective height can portray ratios correctly (a date value twice as large as another will be represented by a prism twice as high), and that readers perceive the height of resulting prisms as ratios.”

    Slocum recommends lightness and hue as visual variables for nominal and ordinal data, and perspective height for numerical data (ratios).

  9. April 27, 2009mark says:

    Hey Bjorn – I’ve not conducted any experiments with prism maps, but I suspect (maybe incorrectly) that map-readers might conflate height or a prism and the volume of a prism…the eye is drawn to the largest volumes on the map, not necessarily the highest prisms…but this is an old problem with choropleth too.

    The one thing that seems odd here with this quote: There is lots of evidence that folks can work very well with estimating differences with lightness, and it, along with size (which is what height on a prism map is) are widely considered the best visual variables for ratio-level data. I’d have to re-visit How Maps Work…MacEachren has a very comprehensive review of the efficiency of various visual variables.

  10. May 6, 2009roland says:

    Hey Mark,
    thank you very much for this blog post.
    It’s absolutly in time, because I am doing a project at the moment where I thought of exactly the same problem.

    Just a few thoughts on this:

    I think the Nepal-Quiz you are showing to argue your opinion is definitly hard to solve. But I think we can do better. Maybe by adding a better color-pattern that seperates values clearer. Except the other facts you mentioned (only half of the data etc.), wouldn’t it be as hard when visualize this situation with a 2D map with exactly the same colors? Without adding labels etc.

    And what do you think about putting the extruded countries away from a global surface and put it on a plane? Then you would have the capability to watch it from the top (same thing you can do with 2D maps) and to move the “camera” in space if you like and understand the extruded visual-language.

    I think another problem with maps showing not area-related but value-realted data is the false information which is caused by the “area of a country”. Same value for example for america and cuba wouldn’t seem the same when encoding it only out of the color without numerous expressions.

  11. May 7, 2009mark says:

    Roland – glad you liked the post. The problem with adding better color pattern, as you suggest, is that it only underscores how little use the height is in reading the map: if you have to rely on color to make the map legible, why not simply use a regular choropleth map in the first place?

    Putting the prism maps on a plane and not on a globe would solve some of the problems (inconsistent scale, only see half the data) but it probably be best to do some user testing to see what works – prism map versus colored prism map versus choropleth versus proportional symbol would make a very interesting study/experiment.

    I agree that one of the weaknesses of choropleth maps is related to unequal sized polygons (your US/Cuba example) – folks have long worried about this issue – which is one of the main arguments for using proportional/graduated symbol maps: If Cuba, even though it is small, had a larger number than the US, it would have a larger symbol (usually circle).

    Cheers

  12. May 26, 2009indiemapper blog » How indieprojector shaped the world says:

    […] us that a Greenland-on-steroids projection and an obscuring, variable-perspective 3D globe are probably not the best for thematic maps. Our math and drawing hurdles here don’t address the much larger challenges of bringing map […]

  13. July 2, 2009DCA says:

    I’m inclined to agree that a lot of uses of GE can be style over substance. I’ll be
    immodest and offer something I’ve done (displaying earthquakes) as an area in which the capabilities of GE make a real difference: http://igppweb.ucsd.edu/~agnew/udq/

  14. September 8, 2009Same same, but different » Blog Archive » Virtual Globes are a seriously bad idea for thematic mapping says:

    […] Finally someone speaking out loudly and describing what I think since the appearance of graphs and thematic maps superposed with GoogleEarth. It’s just nonsene. Too much noise. It’s “against Tufte”, one could say as well. Anyway, read it here. […]

  15. March 4, 20105 best visualizations of 2009 « Geography 970 says:

    […] others. I think my favorite on the list is the Open Street Map animation and I happily put aside my concerns about 3D globes since the point is not any one piece of data…it’s the qualitative impression left by […]