I hate taking strong positions. There are no strong positions that are correct!* Here’s one stated on the ol’ social media circles today:

beautiful data visualization < effective data visualization**

Give it some thought. No really, do. Here’s the real deal:

About 90% of the time a beautiful data visualization is an effective data visualization. If it’s beautiful people are going to look at it, which gets you almost fully to the “effective” state. I realize that’s a bit cynical, but really, just getting people’s attention is the hardest part. I’m thinking 10% of the time you might have a beauty of a map but it tells the wrong story, in which case it’s not just ineffective, it’s misleading at best, unethical at worst.

However, I truly believe that the opposite situation favors the visualization side of things a bit more. Therefore, I’d be comfortable saying that 80% of the time an effective visualization is beautiful (rather than 90%). I’m going to get flack for this, especially as a teacher of effective design, but truthfulness requires even myself to admit that there are times that less than attractive visualizations are still effective.

This is especially true for visualizations of NEW data–data that nobody has ever mapped before. These can get away with a bit of ugliness just by their sheer inventiveness. Here’s a great example from this week:

I tweeted this morning about a new visualization of data surrounding the purchase of BitCoin in various parts of the world. I tweeted it because the subject matter and the implications are fascinating. I didn’t tweet it because of the visualization being “pretty”–which it most assuredly isn’t. A map of something more typical, say population movement, I would’ve wanted to look better. But this one was so new and unique that I was willing to put aside thoughts of the rudimentary map style and overlapping text.

If you want to maximize your effectiveness, though, it is better to err on the side of design: fonts, colors, text placement, line width, projection knowledge, and the like, since there aren’t going to be many times that you get the chance (or think of) such a great new visualization that allows the rest of us to momentarily forget our taste.

*irony

**no disrespect meant, ht @Option_Explicit

In “Some Truth With Maps: A Primer on Symbolization & Design,” by Alan M. MacEachren, the various means of mapping a variable’s intensity are listed and described. Most of these are relevant to mapping today. I’ve included my interpretations and modern updates to these below. I’m using my usual method of illustration here–map slices–just so we get the essence of the idea rather than getting caught up in the map itself.

Size map readers assume that large symbols denote high values, medium symbols denote moderate values, and small symbols denote low values. Don’t mess with the reader’s expectations on this. Symbol size differences, as long as there are only a few, are easily differentiated by the human eye. This makes this form of mapping fairly ideal. It’s pretty obvious, but only use symbol size differentiation for ordinal data. The example below is a recent implementation of the concept, utilizing hexbins to query and display Walmart location data.

Shape Shape is used to denote type, or nominal, data. Because shapes are harder for the human eye to distinguish, the map maker must ensure that they are large and relatively few in order to enable the map reader to more quickly decipher the map. The argument is that a map with a lot of small shapes to look at is not going to be very effective since it will take too long to understand. Icons to depict place-type is probably the most common implementation of shape-based map design. Here, Matt Stevenson’s trail map has only a few, large, icons.

Color Value The lightness of a single color is varied to denote the magnitude of an ordinal variable. This is most useful for heat maps, where the map reader doesn’t need to check the map legend to determine an individual value, but rather only needs to know where the general trend is high or low. With choropleth maps, however, where the reader may want to check the legend to see what each shade means, color value is less useful. More than five shades and it’s too difficult to match the color to the legend key.

Color Hue Nominal data is well suited for display via different colors as long as (1) audience color deficiency is accounted for and (2) the features are large enough to enable the colors to be adequately seen. MacEachren warns against using color differences for ordinal data, saying that people don’t understand logical order of hues (as indicated by the order in the electromagnetic spectrum). However, I differ with him on this point, as I believe that a diverging color scheme–two colors that vary in intensity–can be effective. Blues and greens are common in diverging schemes and are fairly easy for the modern map reader to decipher. This dot map of septic system locations colored by age uses two colors: yellow denotes the youngest age and dark orange denotes the oldest.

Color Saturation The purity of a color is its saturation. This is easily confused with value. The most highly saturated colors (e.g., pure red, green, blue) have medium values. A color palette with a lot of muddy looking colors is a low saturation palette. The infamous fruit palette that we used to see a lot of in GIS maps is a high saturation palette. Depicting the magnitude of something based on saturation will result in a palette of a single hue that ranges from grayish to the least gray. The heat map below is part of a kernel density analysis of aged septic systems.

Texture Maps that use texture to differentiate types in area data are thankfully few and far between today. The fact that it is listed in MacEachren’s book is due to its vintage–when we could only publish maps in black and white we had to use texture a lot more. That didn’t mean it was good, just that it was one of the few options. I always say that if you have to use texture, use it as sparingly as possible by confining it to categories that have the least number of areas or the smallest areas. Data that often requires the use of texture includes geology and soils data, simply due to the fact that there are too many types of things to distinguish to rely on color hue alone. This older map of Yosemite does an okay job of it.

Arrangement Pattern arrangement can indicate type. Different line/dash patterns can differentiate horse riding trails from bike riding trails, for example. MacEachren points out that area data can have a different arrangement of, say, dots, in different areas to distinguish them. Honestly, I’m not seeing much use for this and while I think that density (discussed later) is an important graphic variable, the use of arrangement is antiquated. The example below, therefore, is just a very simple road vs. trail map.

Orientation While we don’t see this too often, it’s by far one of the most visually stunning effects that we can do with our data. MacEachren states that humans are able to quickly and effectively decipher orientation differences in symbols. I feel that orientation is best reserved for data that has a real orientation to map (e.g., flight routes, ship traffic). I assert that there is no need to differentiate non-directional data with orientation as more suitable methods exist. For example, MacEachren’s book has a map that differentiates oil, lumber, reactor, and coal point locations with variously oriented bars when it could be more easily accomplished with hue or even shape. Ocean surface currents, however, is a dataset that has innate orientation, as in the example below, and is particularly suited for orientation-based symbology.

Focus The fuzziness of the features can be manipulated via fade-outs to create a contrast between sharp and imprecise that can indicate unambiguous to ambiguous, respectively. While MacEachren lamented that focus hadn’t made an appearance in many maps at the time of his book, we see it more and more on digital maps now, and with great effect. This canopy height map shows nothing for areas that have (presumably) low or no canopy.

As you can see, in practice, cartographers very often employ more than one of these techniques at a time. Quite simply, if multiple techniques enhance clarity rather than muddy it, then they work.

Density MacEachren doesn’t list density in the book, but it can’t be left out today. Dot density mapping, where one dot is equal to a percentage of what’s inside an area, or even where there is a 1:1 relationship between the dot and the variable, is used a lot today. These have implications for method in terms of how many dots are grouped and at which zoom levels, for digital interactive mapping. Each implementation will differ depending on the data and the interpretation of the cartographer, and most importantly: the size of the dots.

Extrusion Also not listed in the book is extrusion, which is the 3D depiction of a variable via “height.” The early implementations of extrusion weren’t very appealing. Software could only extrude simple features, for one thing, and often those features wound up obscuring important data behind them (e.g., city population maps with sky scraper symbols for each city). However, the cartographer can fine-tune this now and make it work for some data. If the data is suited for it, it is a great technique, since people readily grasp the idea that a taller feature denotes more of the quantity.