By Ellie Fields 19 Mär, 2009

There are few ways of presenting information that are as elegant and as well-suited to their purpose as maps. And because map visualizations immediately place data into a known context, like the map of a country or a state, they are appealing and easily understood.

But there is trouble in map visualization paradise. Geographic data can be misrepresented just like any other kind of data. And one of the most common distortions is one committed by Mediacloud, a Harvard University project, in this map of media coverage.

Harvard mediacould mapping

This cross-border viz infraction caused the Interpol division of the Viz Police to swing into action. Tableau Software Engineer Dirk Karis asks the question:

What country does the BBC cover most?

That viz makes it look like US is #1, Australia (or maybe Russia ) #2. I’m pretty sure the correct answer is the UK, but it’s really hard to see. A Tableau circle plot would make that jump out at you.

What’s going on is that the data (in this case, the amount of coverage) is encoded in color (darker green means more coverage) and plotted on a map. So far, so good. But then the outline of each country is filled in with that color, which all of a sudden makes large countries seem to have higher values than smaller countries. To Dirk’s point, the US is large and dark green in the BBC plot, but the UK is tiny and dark green. Which gets more coverage? Based on the visual presented here, most people will think it’s the US.

This isn’t just an academic question at Tableau. We build our geographic coding to use points, not filled maps, and we often get questions about adding filled maps as another option. While filled maps work well when the data represents densities, on all other occasions a filled map distorts the data. With a point you can more easily compare data points without having the view confounded by size of a given geography. Here’s an example:

a better way to map data

As Tableau presents geo data, big states don’t get preferential treatment. Rhode Island and New Hampshire still matter. Live free or die!

The viz police are forced to cite Harvard’s mediacloud project for distortion of data in the third degree- that is, we give them credit for the infraction being unintentional.


People have been warning about this for a loooong time now. Check out
K. W. Haemer. Area bias in map presentation. The American Statistician, 3(2):19–19, 1949.

Covering the graphic under discussion with the "Viz Police" badge makes it difficult to evaluate that graphic. The border on the image and the obvious layers (looks like a screenshot from Photoshop) make it worse. All in all, the badge tends to create a feeling that the graphic is really, really bad. In other words, it re-enforces your stance that the graphic is bad, through no cause of its own.

I do disagree with the use of points - circles in this case rather than area-less points. It has been well-studied and documented that our visual systems are not geared towards understanding area so circles suffer from the same problems as other areas, it is very difficult for us to accurately judge the value that is encoded by the size of the point.

Another problem I have with the circles, and this is my empirical opinion only, is that points do not relate well to the map. They seem to 'float above it' with either no or a misleading relationship to the underlying geography.

It's worth noting that at least some of the points in the Tableau graphic above do have a white border which, where this is obvious, over-emphasizes those points compared to others that either do not have a border or the border is effectively the same color as the background (maybe it's just my old eyes)

You're merely comparing the color of a country against the colors of the countries near it anyway. The human eye fails at absolute measurements.

Do you plan to add filled maps any time soon as a feature?

Thanks for the comments.

@omomyid: I don’t agree that using areas for measurements is always bad. When all areas are relative to a common reference point, as in the circles, they let you see differences in the data. But when the areas are distorted by quantities that are not being compared, such as the areas of countries on a map, they make comparison difficult.

Here we used both color and size, and focused on the color with the legend. But we could have left size off altogether for a better viz.

@Noah: not at this time.

I agree that using filled circles is distorting the data less if there are areas of different size in your map.

However, I'm wondering how you determine the size of your circles. A circle that fits into the area of the US on a world map easily covers half of Europe if you plot it on top of the UK. Any ideas?