A few weeks ago I was graciously invited by Jeff Howarth to speak to cartography and geography students and faculty at Middlebury College, Dave’s alma mater. I showed some of the work we do at Axis Maps, described our processes, and offered my perspective on what web cartography is all about. The topics were mostly aimed at undergraduate cartography students who may be considering a career path like ours. (While we’re at it, check out some of the student maps.) This post is not at all verbatim but more or less sums up what I said.

The “what do you do?” exchange is always fun for me when meeting new people. When I tell people I’m a cartographer, two reactions usually occur. The first is something like “wow, that’s so cool! I’ve never met a cartographer!” (Lesson: maps make you popular at parties.) Then follows something along the lines of “so what does that mean, like Google Maps?” I then attempt to explain succinctly that yes, sometimes it is kind of like that, but no, it really isn’t.

It’s a little amazing that it’s only taken six or so years for the popular conception of a map—or at least a web map—to become so strongly tied to one type of map, and one exemplar at that. It’s both a blessing and a curse for a practice like ours at Axis Maps, in ways that I hope will be evident as I summarize the way we approach interactive web cartography.

THOROUGHLY DELIBERATE, PURPOSEFUL DESIGN

I made a bad map a couple of weeks ago. It showed 24 hours of bus GPS tracks in Boston, colored according to speed.

Cartographers, trained in their science, would tell me it’s a bad map. It’s a totally inappropriate color scheme for numerical data. It doesn’t generate any clear insights. But the map’s intended audience—the people for whom it was designed—speak differently. It’s eye-catching and novel, it’s reasonably popular, and most importantly it prompts interest and discussion on the state of transit in Boston. Rules and conventions shouldn’t be ignored to the point of misleading or misinforming map users, but just as with wholly “correct” and “useful” maps (which we also try to make!), this particular map successfully accomplished its purpose.

The point is something that seems to define our work and, I think, modern web cartography beyond the general practice of “making maps”: it’s all about purposeful design. Cartographic design is more than visuals and aesthetics; there’s room for the cartographer’s design decisions at every step between the initial earthly phenomenon and the end map user’s behavior.

Daniel Huffman has argued for the human element in cartography with regard to the discipline’s artistic side, and the more I think about it, the more it seems that this is not just about art in cartography; it’s part of what makes a Cartographer something more than a mapmaker. Cartography is about the careful thought behind the design of a map, not just any work (automated or otherwise) that results in a map.

DATA → DESIGN → CODE

So how does cartographic design play out at Axis Maps? We like to think of a project as three-stage process. We begin by finding out what the client wants mapped and for whom, and then assessing and obtaining the necessary data. Next we develop designs for the map, user interface, and interaction based on the known goals, assets, and restrictions. Finally, in a stage that is labor-intensive but conceptually trivial, we write code to build the map as designed. Without getting too far into the boring details of how we work, I want to mention a few notes on each stage.

Data

Anyone who has tried to make a map, chart, or anything like that will know that working with data is an easily underestimated task. Data come in a million formats and are often messy. Jeremy White, graphics editor and cartographer at the New York Times, has said that when people ask his advice on what software to know for his line of work, to their surprise he answers Excel. It takes at least passing familiarity with a variety of formats and scripts and tools to be prepared. And getting data onto a map isn’t just a matter of using ArcGIS anymore. I haven’t used ArcGIS even once in the past four years.

I’ll say two specific things about data. First, we always take care to obtain a data inventory from the client and to develop a data model early on. The data inventory (a list of everything that needs to be shown on the map) is an important first step before we begin designing anything, because obviously we need to have complete knowledge of the requirements in order to come up with a good design. Similarly, the data model (the way the data are organized, basically) will be necessary to know how to write code that loads and processes the data later on.

Second, all of that matters because complexity of the data and map can vary a lot, and it can’t be unknown when we go to design an interface. The chart below, from a paper by Robert Roth and Mark Harrower (PDF), explains why complexity matters. (It’s talking about interface complexity rather than data complexity, but we find them to be related.) We need to know about complexity and the map’s audience in order to execute a successful design.

Design

If there is one clear thing I can say about our design process, it’s that it works like this: mock up EVERYTHING. Everything! We try not to leave anything to imagination. We generate mockups for every interface state, every map view, and every interaction. This usually means a couple dozen screens in the end, showing a step-by-step simulation of a user interacting with the map. We think that locking down all these designs before writing a single line of code is crucial to smooth development and good design. Otherwise we run the risk of cobbling together designs on the fly while writing code, resulting in a messier product.

We’ll always miss a few things, but with enough thinking and discussion we manage to identify most problems before encountering them during development. Our design process, like most I’m sure, is very iterative and involves a lot of attempts and review. Ben, our main Design Guy, draws on experience, conventions, constraints, user feedback, a keen sense of aesthetics, and, I assume, magic to turn ideas into great-looking and smoothly functioning designs. (Maybe he’ll have a chance to describe his methods here sometime.) He notes that there are always a zillion ways to attack a design problem, and for every alternative there is always a better one. We discuss to death possibilities for every little detail until the optimal solution is achieved. Ben’s idea of improvement in design skill is being quicker and requiring fewer attempts to arrive at the best solution to a problem.

Code

Writing code takes up the bulk of our time, but in concept it’s almost a formality to us. It’s all about choosing the right tools for the job (Flash, OpenLayers, Polymaps, jQuery, and so on) and then building what we’ve already so carefully designed. We don’t do this work in order to do interesting or novel technological things; we do it to make good maps. If cool technological developments come out of it, all the better, but it’s almost never the main purpose. In my own invented definition of cartography, cartographers are not the ones whose drive is to develop mapmaking technologies. Another related community does that, spending less energy on designing actual maps. It all works well as long as the groups exchange knowledge and each knows what the other is doing.

Our coding process goes something like this. 1) Load the data. 2) Make things work. 3) Make things pretty. Like I mentioned before, having everything designed ahead of time is vital. We can start with something rough but functional without worrying about design, because we already know how it will look and behave in the end. It also lets us know when we’re finished; interactive projects have a way of never ending if there are no clear goals at the outset.

After hearing from enough of my cartography peers whose hatred of programming burns with the fire of a thousand suns, I must say this: yes, coding sucks. I write code all the time, and it often makes me want to punch the computer in the face. But it’s worth it. Totally worth it. It only takes a little skill to produce awesome things. A willingness to write some code opens a lot of doors, and it doesn’t require devoting a lifetime to becoming a master programmer. It doesn’t even require being a good programmer. It’s just another skill, not so different from, say, drawing Bézier curves in Illustrator for static work. Nathan Yau’s tale (and his Visualize This book) is a good one to learn from for those who have resisted getting into programming.

WHERE DOES DESIGN BEGIN?

After describing the design we do, it’s worth noting that visuals and user experience design are only one part of the overall process of designing a map. Kirk Goldsberry, visiting scholar at Harvard and professor at Michigan State University, recently impressed upon me that design in a cartographic context—broadly meaning the decisions that go into map—is not merely figuring out the visuals, but rather exists in the entire mapping process, something I touched on earlier. Leaving out map use for now, consider the progression from phenomenon to graphic. At one end is the actual thing that is happening, at the other end is the map that represents it. In the middle are data, meaningless in isolation and not to be confused with the phenomenon itself.

Above are some activities that exemplify the progressions from phenomenon to data, from data to graphic, and the whole thing from phenomenon to graphic. Ideally a cartographer designs the entire process: what data are collected, how they are collected, how they’re organized, how they’re represented, how the map looks, how interaction works, &c. Dr. Goldsberry gave the example of old-timey explorers. They went places, recorded the data themselves for the purpose of making a map, and then they crafted the map itself. They designed everything. Sometimes a cartographer can still own the whole processes, but it’s rare these days, especially in web mapping. Realistically I think most activity falls either between phenomenon-to-data or data-to-graphic, with most of us who call ourselves cartographers existing in the latter category. We work with the data we have, but it’s worth bearing in mind that we’re doing something (i.e., making a map) that the data may not have been meant for, and this can affect our user experience design decisions.

WEB MAPPING, or PUTTING THINGS ON TOP OF OTHER THINGS

Returning to Google Maps, it has defined not only the layperson’s idea of a web map but also the web mapper’s idea of a web map, it seems. Ever since the early days of Google Maps mashups, the trend in web maps has been basemap + stuff on top. There’s almost always this strict separation of layers—layers that often were not designed to go together, although that part is gradually on the decline. We’ve advanced to the point where pretty good cartography is possible and easy in this framework (thanks to tools like TileMill), but it remains the case that web cartography usually means designing around the tiled Mercator slippy map system, and often using someone else’s tiles, instead of seeking the ideal solution. We all do what’s feasible within technological and time constraints, of course. At Axis Maps we take advantage of the built-in capabilities and user familiarity with standard tiled web maps all the time. But I do sense a risk that “web map” is coming to mean only one type of map, the “things on top of other things” map.

So perhaps my purpose today is to remind us all that there’s more than one kind of web map. Cartography is not Google Maps. It’s not OpenStreetMap. It’s not mashing up geotagged data from various APIs. It’s not rendering tiles. It’s not “geo” (“geo” is a stupid non-word and I wish it would die). It’s not GIS. Cartography is in the thoughtful design of maps, no matter how they are built or delivered.

Today we’re pleased to show off a pet project that’s been occupying us off and on for nearly two years. After some emotional separation issues, we are declaring finished a few typographic map posters—one of Boston, and color and black and white flavors of Chicago. Everything in these maps is made of type.

These look good hanging on a wall, so of course prints are available. Check out the page we’ve set up with some more detailed images and links to get copies for yourself.

I began this project with the Boston map, thinking it would be fun to expand the style of my small party announcement map to a full city. The idea caught on here at Axis Maps and soon Mark and Ben had parallel effort underway for a map of Chicago, a city to which several Axis Mappers have some affinity. Ben took the lead on that map, and some twenty months later we both added our respective finishing touches and reluctantly let go.

There was nothing automated about making these maps, unless you count copying and pasting. Everything was laid out manually, from tracing streets over an OpenStreetMap image, to nudging curved water text, to selectively erasing text to create a woven street pattern. The Boston and Chicago maps differ in style, but the end result is similar: from a distance it can appear as an accurate reference map, and as you get closer you notice the thousands of words it comprises.

This has been a fun, if long, process, and we hope other people can enjoy these maps as much as we have. There are only two cities for now, but look for more in the future! Our list right now is San Francisco, New York (Manhattan), and Washington, D.C.

Recently, we took on a nice little print mapping project for a few hotels located in downtown Madison, Wisconsin. The project involved making a one-sided, page-sized map showing hotel locations and the locations of a few points of interest in the area. The idea was that hotel guests could use the map to find their way around downtown as well as get a sense for where they were staying in relation to the university, interstates, airport, etc. The map was to be printed in grayscale, plus 3 spot colors (red, yellow, and blue).

Before starting out, we discussed the possibility of sharing the project with those interested in seeing all the stuff that goes into designing a map like this. The map design process is notoriously difficult to articulate and we’re keen on the idea of making pieces of it more transparent, where possible. One option was to screen capture the hotel map as it appeared in the production software at regular time intervals from blank page to finished product. So, here is a sequence of 116 images, originally captured at 10-minute intervals, compiled to show the evolution of the hotel map in just under 2 minutes. Clearly, not all maps are made in the same way, but this should expose some of the kinds of design decisions made in a relatively simple project like this.

Watch the larger version of Map Evolution (990 x 766px) — best for seeing change in map details.

After posting our election map last month, we received a number of excellent comments and suggestions. It’s late, but I thought I’d finally post the couple of variations of the map that I’ve managed to find time to put together. The maps below do two things differently from the original:

• Vary the brightness of counties by population density rather than total population. This was a frequent suggestion. I think it has a few of its own drawbacks too, but it looks pretty good.
• Different color schemes. Just for fun, I’ve used the purple color scheme that has become common in recent elections. I also liked the suggestion in one comment to saturate colors by margin of victory, so I’ve done that too. In these, full blue would be total Obama domination (Obamanation? Obamadom?), full red would be the same for McCain, and gray is an even split.

No snazzy posters this time. Just a few map snapshots.

First, the original colors mapped by population density, as posted in the comments on the original post.

The purple color scheme. First by total population:

And by population density:

Margin of victory by total population:

Margin of victory by population density:

Apologies for any trouble seeing the images. It’s tricky to find a brightness that will look right on every screen.

When we were building GeoCommons Maker! one of the key map design challenges we faced involved producing semi-transparent choropleth maps. Choropleth maps are perhaps the most common type of thematic map and are regularly used to show data that is attached to enumeration unit boundaries, like states or counties. Ever seen a red state / blue state election map? This is a basic choropleth. There are a lot of more sophisticated ways that choropleths can be made to best represent a given data set, for example, by playing around with classification, categorization, choice of color scheme, etc., but we won’t get into those here.

I want to talk about color. Traditionally, choropleth maps are read by looking at the general pattern of unit colors and/or by matching the colors of specific map units to a legend. Other reference data is often removed from the map because it is either, 1) not necessary to communicate the map’s primary message or 2) makes communicating this message more difficult. It could be argued, for example, that other reference map information, like green parks, gray airports, brown building footprints, and blue water distract readers from seeing the general pattern of choropleth colors on the map, which is where the map’s most important message can be found.

For GeoCommons Maker!, we wanted to allow people to make a kind of hybrid, semi-transparent choropleth map that would show both thematic data (colored choropleth map units) AND the rich reference information on popular map tiles (e.g., Google, Microsoft Virtual Earth) without sacrificing map reading and interpretation ability and confidence. We believe that there are lots of times when reference and thematic data can work extremely well together to really benefit a map’s message (e.g., a soils map that shows terrain or a vegetation map that shows elevation). So, we wanted to build this functionality into Maker!, and allow people to make maps that show the best of both worlds.

The Problem with Transparency

The fundamental problem with transparency is that the color of semi-transparent map units can shift due to the visibility of color that lies beneath them. This is not at all surprising, but can make the basic legend matching task difficult, obscure the pattern of color on the map, or just as bad, make patterns appear out of nowhere. Here’s a look at what happens to colors using the same semi-transparent choropleth map units on different backgrounds. These are screen captures from early design mock-ups for Maker!.

The first image shows (hypothetical) opaque choropleth map units with a 7-class color ramp. The next three images show the same units at 50% opacity on top of Google terrain, streets, and satellite imagery. Notice how colors shift when compared to the opaque map at top? See how lightly colored units nearly disappear on the streets map, and darkly colored units nearly disappear on the satellite map? Yikes!

Mock-up of an opaque, 7-class choropleth map for Maker! (Google terrain)

Same mock-up, at 50% opaque (Google terrain)

Same mock-up, at 50% opaque (Google streets)

Same mock-up, at 50% opaque (Google satellite)

The Solution to Transparency

We employed three design solutions to ensure that semi-transparent choropleth maps in Maker! would work, despite potential map reading problems: 1) unit boundaries, 2) data probing, and 3) transparency control.

1) Unit boundaries. In Maker’s choropleth maps unit boundaries are color coded but remain opaque, even when unit fill color is semi-transparent. This gives map users some true color information to work with, and should improve their ability and confidence to spot map patterns or match colors to a legend. In other words, while unit fill colors can get you close, unit boundaries can get you the rest of the way there.

Screen-shot from Maker! showing opaque choropleth unit boundaries

Corresponding legend for the above map

2) Data probing. We also took advantage of a relatively common and very helpful interactive map feature known as data probing. Exact values for any choropleth map unit can be obtained by clicking on them. In Maker!, we designed the data probing feature to go one step further and give values for all of the possible attributes associated with each map unit, not just the mapped attribute alone (see the scrolly list, shown in the probing pop-up below).

GeoCommons Maker! Data Probe

3) Transparency control. Finally, we gave mapmakers a transparency control, as well as a chance to take some responsibility for how well their maps communicate. The transparency control lets mapmakers decide what works and what doesn’t. Given the huge range of possible maps that can be made with Maker!, some user controls like this are necessary (as well as being kinda fun!). Here, transparency can be adjusted for a custom fit with any chosen tile set, color scheme, or other mapped data. Settings on the control (shown below) range from 50-100% opaque.

Screen-shot from Maker! showing the transparency control

The Best of Both Worlds

Our decision to include semi-transparent choropleth maps in Maker! should give mapmakers and map users the best of both worlds. A semi-transparent choropleth is truly a hybrid map in that it can potentially offer all the advantages of combining rich reference data (i.e., underlying tile sets) with great thematic data (i.e. overlying choropleth units). Hopefully the choropleth maps coming out of Maker! will be easy to read and good looking, too!

The first time I used Google Maps I knew that the world of cartography had just gotten a lot more interesting. It blew my mind. What really struck me then (and still does today) is that I didn’t have to learn how to use it: It just worked. It didn’t come with a manual, and I didn’t need a class in it. Rather, I would think, “Hey, I wonder if…” and sure enough it did just that. First try. It worked. What was happening was that my expectations of the map and the feedback it gave—and the speed at which it gave that feedback—left me feeling empowered to explore more, rather than frustrated or confused.

This is true of all the best tools in our lives: they make us feel confident (even smart), not intimated or confused or frustrated. And they quickly become completely transparent. The master violinist, photographer, or painter all work so comfortably with their tools that they are able to translate their powerful and nuanced intentions into a physical reality. We’ve all experienced this – when the interface between our cognitive and emotional selves and the world around us disappears and we are able to lose ourselves in a great book or a great movie (you cease to realize you’re sitting in the theatre watching reflected light on a screen or scanning printed characters on page).

When our tools disappear and become transparent, we are at our best. Psychologists call this ‘flow’ and I think it is the singular defining state of creativity: it’s what happens when we are so deeply engaged with the work we love that we lose track of time and need to be prompted by loved ones to stop and eat occasionally. Mozart had this problem, Newton had this too, and to a lesser extent, so do I every time I fire up Google Earth.  I often joke that Google Earth should come with a warning label: You will be here happily for hours. Proceed with caution.

Now reflect on how rare that experience is in the world of software and web interface design. Why is that? Why are we content to create maps that merely don’t crash? Below I outline how we might, as designers, aim a little higher.

The problem is, as design guru Donald Norman points out in his classic must-read The Design of Everyday Things, most people expect to be flummoxed by new technology, that’ll it be hard to learn, that it will be unpleasant at best.

Why else would people refuse to upgrade software despite obvious problems with their current version — because they’ve done the math, and the current flaws are better than having to learn a new version. Indeed, most people blame themselves when something doesn’t work, saying, “I must be stupid because other people know how to use this,” when in fact it’s most likely a counter-intuitive UI and poor or missing affordances that are to blame. The only reason why other folks know how to use it is because they’ve learned, through trial and error, how to work around those flaws. If software elicits “That made no sense, but I guess it worked…I’ll try to remember that for next time,” it is badly designed. If it elicits “I bet I can do x by using y,” you’ve earned your paycheck (ironically, most payroll systems I’ve seen are horrendous).

How do you know this isn't your interactive map or web site?

Success in UI design is not measured by “Did the person get something at the end?” but rather by “Did they grasp what the tool was capable of and how to use it quickly? What was the mental and physical workload required? How many dead ends did they go down before they found success? Did they enjoy the experience?” among other important questions. The TLX scorecard (designed by NASA) and the GOMS test are two such approaches used by savvy designers to score how well people use their tools, not merely if they can use them (or, as we often see, only use them if they’ve taken lengthy training courses).

Case in point: A Big Ten university I know reserves a mandatory full afternoon to show every new employee how to use the phone message system, representing millions in lost productivity. Uh, Houston, I think you might have a problem, and it’s not your employees. This isn’t just design snobbery it’s massive waste of money and time.

THE IMPORTANCE OF FLOW

Let me propose that the best user interfaces (UIs) become transparent because they engender ”flow.“ For me, this is the holy grail of good design. The 9 components of flow, based on Csíkszentmihályi’s work (see his TED Talk) in the 1970s, can be used as a scorecard for any UI we design or use:

1. Clear goals (expectations and rules are discernible and goals are attainable and align appropriately with one’s skill set and abilities).
2. Concentrating and focusing, a high degree of concentration on a limited field of attention (a person engaged in the activity will have the opportunity to focus and to delve deeply into it).
3. A loss of the feeling of self-consciousness, the merging of action and awareness.
4. Distorted sense of time, one’s subjective experience of time is altered.
5. Direct and immediate feedback (successes and failures in the course of the activity are apparent, so that behavior can be adjusted as needed).
6. Balance between ability level and challenge (the activity is neither too easy nor too difficult).
7. A sense of personal control over the situation or activity.
8. The activity is intrinsically rewarding, so there is an effortlessness of action.
9. People become absorbed in their activity, and focus of awareness is narrowed down to the activity itself, action awareness merging.

HOW DO WE GET THERE?

There is no simple magic formula, I suspect, for what defines a good interactive map (or any user interface) but I think some salient qualities include the following:

1. Give Them Something Quickly. Nothing boosts confidence like finding the power button immediately or sliding your credit card in the reader the correct way the first time (aside: omg card reader designers— after 30 years this is the best you can do? Seriously?!!). I’ve always liked Apple’s “welcome” movie that plays the first time you start up your new Mac – all you’ve done is find the power button but already you feel like you’re off to a good start. A blinking c: prompt is no way to greet your clients.

2. Adapt to the Skill-Level of the User. Maps and software should be smart enough to adapt to the level of the user, from offering pro-level shortcuts to revealing more advanced features as needed — not when they can’t be used and simply clutter up the screen and taunt the user with ”shame you don’t know how to activate me cause I’m all grayed-out.”

3. Understand Affordances. All interfaces, from airport signs to lawnmowers, live or die by them. If you are a designer, you need to eat, breathe, and sleep this stuff for the world already has too many ”Norman Doors.“ Named after Donald Norman (who talks about them in his aforementioned book), these are doors that have a horizontal bar across them and thus suggest that you’re suppose to push them, when in fact you have to pull them (RULE: horizontal bars to push, graspable vertical handles to pull). In the panic of a fire, such design flaws take on new significance.

4.    Eliminate Features Ruthlessly. My two current favorite UIs are the Google Search Engine and the iPod. Both are the model of simplicity and both burst onto the scene and quickly dominated their markets by doing the incredibly counter-intuitive thing: offer the user less. They both removed a whole bunch of features the competition thought were essential. Why? Because we are all swamped with too many choices in our lives (see “The Tyranny of Choice”) and simpler tools are (1) faster to learn, (2) faster to use, (3) cheaper to make, and (4) and less likely to break.

Unlike just about every product ever, over time the iPod has in fact gotten less complicated: The Gen 1 iPod had twice as many buttons as the Gen 3. The iPod shuffle? It got rid of the screen! Heck, the shuffle even eliminated the power jack and let folks recharge directly through the headphone jack (I still don’t know how they did that one). And it sold like hotcakes because it was cheaper to make and easier to use — they killed the features that were underused and kept just the stuff that really mattered to most casual users (pro users can still buy the more powerful models of iPod). And most folks learned, hey, you know, I really didn’t need all that stuff after all, and I just saved a bunch of money…and that is a happy customer.

5.    Build a Well Labeled Emergency Exit. If the user feels like at any moment they might break it, they won’t venture far. The best UIs increase confidence by providing reassuring feedback that progress is being made and encourage the user to keep going. We’ve all been stopped by cryptic warning messages that leave us feeling unsure of whether to proceed or cancel (I usually cancel, unless I’m feeling dangerous or it’s not my computer). Good ideas include a big reset button, unlimited undos, and lots of sign posts such as “Before we close the window, would you like me to save your work for you?” More advanced users can turn off such warnings once they’re weaned off of them. I have archiving software that asks me three times in three ways if I really, truly want to erase a drive and overwrite it. Given the cost of a mistake (10 years of my entire digital life) this five seconds of forced careful thinking is a suitable insurance premium.

Ben's pick for UI of the week: Swedish light switches.

I asked Andy, Ben, and Dave to name their current favorite UIs and Ben suggested these cool light switches in his apartment in Sweden. He writes “(1) really big and square, (2) in the dark, just start smacking the wall, and (3) feels really good to smack the wall in the dark.” Hard to argue with that. Compared to the light swithces we have here in North America these are easier to use (less effort) and result in less fumbling (faster to use, fewer mistakes) and they remind us even the humble light switch can be improved with some careful thinking.

I made mention of 3D pie charts in an earlier post and thought I’d outline exactly why they are such a bad idea. As both a teacher and designer I campaign hard against “chart junk” and the needless and confusing eye candy tricks that software companies create to clutter-up our lives. I know these companies need to offer something to try and convince us to upgrade to the new version, but let’s be clear: drop-shadowing every element on the page, or adding an outer glow to the text isn’t going to make your message any clearer, and will most likely distract from the very thing you’re trying to show. My design philosophy can be summed up as:

In cartography, aim to be clear, not cool.

Anything that doesn’t contribute to the message, or worse, distracts from it, probably doesn’t need to be on the page. Since maps are small and the world is large, every inch on the page and every pixel on the screen has to count and we can’t afford to waste any of them. Draw what you need, and no more. Fans of Edward Tufte, Presentation Zen (recent post on ‘chart junk’), or old-school cartographer’s like J. K. Wright and Arthur Robinson will recognize all of this – this is hardly a new message. But it is one that still needs to be heard, apparently. For a quick overview of many of these arguments I’d strongly recommend reading John Krygier‘s excellent post “How useful is Tufte for making maps” (his 20 Tufte-isms is a great crash-course in Tufte).

Consider this graph (from here):

Pure chart junk: This pie chart commits a half dozen major design mistakes rendered it as little more than visual junk food (looks tasty at first, but isn't that good for you).

As an information graphic, let’s step back and think about what a pie chart is suppose to do and how it works at a perceptual level: Pie charts are used to tell us (1) how much of something exists, and (2) how much that is compared to the other categories. ‘How much’ is encoded by the size of the pie (or segment) and ‘relatively how much’ by the internal angles of the pies and/or their relative sizes. To extract data from the graphic you have to be able to quickly visually compute both areas and angles.

The bad news: Years of testing has shown that most of us are really bad at estimating the areas of even simple shapes – just try visually estimating how much carpet you’ll need for a room, especially if the room isn’t square if you don’t believe me – and we’re pretty bad at eyeballing and comparing angles too.

Not convinced, you say? Looking at the pie chart above:

• What percent of the total does DP Tech have?
• Is that more or less than IBM?
• How much more/less?

Now think about presenting those data as a boring old table: DP Tech 4%, IBM 5%. Done. Simple. Think about the difference in mental workload, and the confidence you have in your answer when the data are presented as a 3D oblique pie chart and when they’re numbers sitting in front of you. This problem is so commonplace (and yet ignored) that most folks resort to putting numbers on pie charts because the graphic itself is not sufficient, which is waste of ink, their time and mine. If you have that little confidence in your charts, just give me the numbers!

Here’s a rule of thumb I like to use:

If a map/graphic needs ‘crutches’ like number labels and can’t stand on its own, don’t use it. It’s the difference between “A Tale of Two Cities” and “A Tale of Two Cities: A Novel.”

People read the simple 2D pie faster, with greater accuracy AND greater confidence.

The same data is presented three different ways above, and each change made to “enhance” the simple 2D pie chart makes it worse because the two basic perceptual tasks – ‘how big is something’ and ‘what are the angles’ – are much harder to perform when the pie is lying down. This is exactly what happens when design decisions are made in a vacuum and based simply on “it looks cooler this way” rather than an understanding what we need from a graphic to make it readable/work.

Problem 1: Adding the 3rd dimension adds no new information to the graphic. That’s bad because it is wasted ink (that could be doing real work) and it requires the tilting of the pie so the designer can show off the 3D effect. If the height (z-dimension) added some additional data (i.e., a second data variable), it might be worth adding, although I would caution against that since we’re even worse at estimating volumes than we are at estimating areas (which is why “how many jelly beans in the jar” contests or “how big a moving van do I need?” continue to challenge us – we’re terrible at numerically estimating volumes beyond the crude level of “bigger / smaller”).

Problem 2: On both oblique pies the scale is not consistent across the graphic. In other words, the same pie segment will look larger or smaller to you the observer simply based on where it lands in the circle…things closer to us will look larger, even if they’re not. This couldn’t be more counter-productive when we’re simultaneously asking the viewer to estimate areas. This is an absolute rule: If you expect people to judge the size of things, don’t change the scale of the objects on them.

Problem 3: Splitting the pie apart makes matters worse because the further objects are from each other, the harder it is to compare them (which is why we like to hold things side-by-side if we want to carefully compare them). Why? The extra time and effort it takes for your eyes to search for and acquire the now-separated segments uses-up your precious (visual) working memory and requires more eye trips back-and-forth to make the same estimation. Cognitive scientists call those back-and-forth trips extraneous cognitive load which cut the available brainpower (working memory) that can be devoted to the real task of comparing segments (intrinsic and germane cognitive load).

Solution? Simple: Stop using oblique, exploded, jazzed-up 3D pie charts. 2D work better, are easy to read, faster to read, and easier to make. Importantly, they also can be drawn much smaller and remain legible – as cartographers we’re always looking for ways to do more with less ink. If your powerpoint slides feel naked without fancy transitions and giant 3D graphics, you’d do better to work on the substance of the talk, rather than bury your good ideas in a pile of chart junk.

I’ll leave with one of my all-time favorite spoofs – the Gettysburg Address as Powerpoint.

Update, Dec. 22: A few variations of the map technique are posted here.

We spent some of our spare time last week exploring data from the 2008 presidential election and thinking of some interesting ways to visualize it. Above is one map we put together.

One thing we sought to do was present an alternative to cartograms, which are becoming increasingly popular as post-election maps. Cartograms are typically offered as an alternative to the common red and blue maps showing which states or counties were won by each candidate, wherein one color (presently, red) dominates the map because of the more expansive—but less populated—area won by one candidate. Election cartograms such as the popular set by Mark Newman distort areas to reflect population and give a more accurate picture of the actual distribution of votes. A drawback of cartograms that we’re very aware of, however, is that in distorting sizes, shapes and positions are necessarily distorted, sometimes to the point of making the geography virtually unrecognizable.

Our map is one suggestion of a different way to weight election results on the map while maintaining correct geography. What we’ve done is start with a simple red and blue map showing which candidate (Republican and Democrat, respectively) won each county in the lower 48 states. Then, to account for the population of those counties (or, the approximate distribution of votes), we’ve adjusted opacity. High-population counties are fully opaque while those with the lowest population are nearly invisible. Against the black background, the highest concentrations of votes stand out as the brightest.

We’ll let viewers be the judge of its cartographic effectiveness, but we hope you’ll at least agree that it looks pretty cool!

Click on the image at the top of the post to view a larger version, or see it in a Zoomify viewer, or download the full size (suitable for printing).