Info We Trust

by R J Andrews

  Except for a pure laser beam, the colors we see are actually ranges of the visible spectrum. Many perceived colors, like magenta, do not exist at any specific wavelength or in any natural rainbow. These extra-spectral colors are like the musical chords of the color spectrum, wholes perceived from physiological and psychological mixtures of light.

  Distinct hues are farthest separated from each other around a color wheel. Complement pairs are the most distinct, they straddle the wheel. Distinct palettes interest us because we perceive color according to difference. From a survival perspective, colors are most relevant relative to their background. Throughout the day, and throughout the year, lighting conditions change the perceived color of objects. We have adapted to be sensitive to difference, not absolute value. Imagine you are hunting for ripe apples against a field of tree leaves. To contrast the red color of desire against other environmental cues, we change our pupil size and photoreceptor sensitivity. This lets us hunt for apples at noon and dusk. We change to make effective comparisons across a variety of lighting conditions.

  After difference, learnability is the next quality of a good color palette. Colors and categories must be mentally linked, remembered, and referred to in conversation without confusion. Palettes should be comprised of hues that suggest familiar names. Intuitive color palettes that reflect some semantic meaning, such as making bodies of water blue, also helps. In addition to mimicking how we experience the natural world, semantic color builds on other palettes readers have seen before.

  Berlin and Kay reported that a set of 11 basic color names appear in all mature languages, usually beginning with words for dark-light, then the four opponent colors (red-green, blue-yellow), and finally purple, orange, pink, brown, and grey. Their findings still resonate with many color enthusiasts, despite being scaled back since original publication in 1969.

  Typically, a categorical palette must be limited to a handful of colors. Hues are impossible to distinguish if they are packed too closely around the wheel. Too many colors will make it difficult to remember which colors map to which categories. Shades of the same hue might imply that one category is a lesser version of another. It is difficult to assemble a palette of many more than six hues that does not run into problems.

  Hue is an attribute of color dependent on wavelength, not color intensity or lightness. The word is originally linked to skin complexion (animal hide shares the same root).

  But we do not always have to be typical. Grouped color schemes can increase the number of color categories beyond a handful. They do this by arranging categories into a conceptual hierarchy. Each chunk of the hierarchy gets one of the main colors. Then, inside each chunk, varieties of that named color can distinguish more. For example, consider this map of my Hudson River Valley hometown. Blues are water. Greens are nature. Warm colors are buildings. Greys are paved surfaces. Browns are crops. It really is a colorful little village.

  A highlight color is a type of category used to distinguish a single class from all others. It is often a more visually striking version of another color, or an alerting color such as red or yellow.

  Color demands respect for how an entire stack of technological, biological, and cultural systems conspire against successful decoding. Our technology is not perfectly aligned with our biology. Biology varies across all readers. Each person's mental processing of color is different. It is easy to miss how wacky the handshakes are between our data, color mapping, the way colors are displayed, and the way they are perceived.

  Take a brightly colored object such as a bright red ball. It is a saturated red hue all over, independent of the changes in color caused by the play of light and shadow upon it. That is, hue and saturation are inherently defined by the object color, not its lighting. Across the shaded surface of the ball, the lightness and chroma will vary. Make the lighting twice as bright, and the ball will appear more colorful and brighter. Dim the lighting and it will become less so. But, the lightness and chroma will remain the same because they are relative to the light.

  MAUREEN STONE, 2003

  The real color spectrum is expressed with different color models for different purposes. The most familiar models help us program machines to display color. RGB instructs screens how bright to make an array of red, blue, and green light sources, like the iPhone diamond matrix here. The CMYK color model dictates where to print cyan, magenta, yellow, and black (key) ink on the page. No print or display technology is capable of producing all of the colors our eyes can detect. Both RGB and CMYK are languages convenient to machines, not to the human eye. How we actually perceive color makes things complicated because everyone sees differently. We are each equipped with a unique set of biological hardware. Incredibly, our brain generates our rich color experience from relatively simple sensory input: long, medium, and short (LMS) cones, named for the profile of visible light wavelengths each is specialized to detect. The density of our eyes' rods and cones varies amongst all trichromats, people with normal three-cone color vision. Some, mostly men, only have two types of cones. These dichromats suffer some impairment in the perception of the red-green spectrum.

  Tetrachromats have four independent channels for conveying color information, a condition present in some animals and studied as a phenomena in a small percentage of humans who are able to detect a greater variety of colors than common trichromats.

  Printing, like this book, can only display certain colors (colored at right). A screen has its own limited gamut (the RGB triangle). Neither can show all of the colors human eyes can see in nature, represented by the area under the arch, from about 450-650nm.

  A color scale should vary consistently across the entire range of values, so that each step is equivalent, regardless of its position on the scale. In other words, the difference between 1 and 2 should be perceived the same as the difference between 11 and 12, or 101 and 102, preserving patterns and relationships in the data.

  ROB SIMMON, 2013

  With only three types of cone, we get far more than just three colors. Our brain's visual system interprets the difference between the three LMS cone signals. We can imagine these signals locating colors across a multidimensional space, similar to how multiple satellites can triangulate a specific position. The combinatorial way we see, from cones to color, has important implications for design. Sensory input from all three cones fuse to make us best at seeing gradations along the dark-light achromatic range (shown as the horizontal axis in the below diagram). Even with full color, lightness is still most important.

  LMS (long, medium, and short) cones are popularly associated with the colors of red, green, and blue, but this threatens to mask how much their ranges overlaps.

  Perceived lightness, called luminance, should be consistent across color palettes. The HCL (hue-chroma-luminance) color model pursues even perceptual distance across its color volume. It arrives by way of more than a century of color science innovation, which included incorporating more computing power and running more experiments where people perform color differentiating tasks, such as determining the “just noticeable difference” between shades and adjusting colored light to make differently painted swatches match.

  The population ratio of LMS cones varies from person-to-person, but we generally have more L than M, and few S. This relative scarcity of short cones impacts our ability to appreciate intensely blue colors.

  The “opponent process” theory of color is responsible for connecting the overlapping spectrums of our LMS cones to how we perceive color. The resulting three color pairs are called opponent, or antagonist, channels: yellow versus blue, red versus green, and light versus dark.

  The eyes we design with are not the eyes of the reader. Designing for human vision reminds me of old graphic design handbooks. They warned: Expect your work to be mangled by photocopiers. As you work with color, simulate how design appears to different color deficiencies and on different devices. No colors, even reds and greens, have to be abandoned. They often just need to be tweaked.
<
br />   Paint by Numbers

  Quantitative color palettes map a change in numerical value to a change in color value. Many palettes vary only one dimension of color. An increase in number, as we saw, can correspond to an increase in darkness. Color models are often represented by a perfect cube, cylinder, or other tidy volume. Digital tools often have us navigate 3-D color space by using the hue dimension as a conceptual backbone and presenting us with the remaining 2-D slice of color fixed to that part of the backbone. Each of these perfect volumes reflect the computer-centric regularity of machines and their controlling mathematics. Some creative palettes map data across a spline that cuts through a color volume, varying color in all three dimensions as values change. Through experiments of how people actually see, we have learned that the actual shape of our perceptual field is quite irregular. Human color space is lumpy.

  The color wheel's full circle, with no beginning and no end, is a strange friend for data. Indeed, quantitative rainbow palettes introduce a conceptual paradox. Data minimum and maximum values should be visually distant because they are numerically distant. Rainbow max and min, however, are colorfully right next to each other. Using the full rainbow is like exiting a tunnel and finding yourself back at its entrance.

  Color transition points can help indicate meaningful thresholds within a data range. Diverging scales direct attention toward extremes by using a neutral color for the hinge value. The color legend at left emphasizes height above and depth below sea level. Quantitative colors do not always need to be evenly distributed. The example below allocates a warm palette strategically. The only data of interest, the top 10 percent, is highlighted, while the bottom 50 percent is colorfully weakened. This intentional palette stretching blends the benefits of qualitative and quantitative approaches.

  Sometimes, a fluid color spectrum is created by interpolating hundreds of color variations along a color ramp. Our ability to actually differentiate between noncontiguous colors is poor. It hovers somewhere around three or four shades of saturation or luminance. Paradoxically, dividing a rich spectrum of hundreds of colors into a few discrete bins can help people discern more colors. Go ahead and use a color ramp for aesthetic reasons, but know that it is at the expense of decoding information. Instead, bucket (or step) quantitative palettes into discrete (or quantized) bins.

  More colors is not the same as more discernible colors.

  Total Color Challenge

  Color can be defined numerically at so many different stages: wavelength of light emission, absorption and reflection, design mapping, red-green-blue pixel encoding, our eye's reception, and our mind's perception, which takes into account complex factors such as how nearby colors interact and the relative size of each area of color. Numerical systems help us produce color, but none of them perfectly match how we think about color.

  Perceptual effects [can] strengthen or destroy a design.

  MAUREEN STONE, 2003

  How colors are perceived, not how they are generated, is what matters most. Successful color decoding happens when we account for the reader's eyes, mind, and experience. We want color to support a successful encoding-to-decoding chain, but we also want color to help make our design look good. These two goals can create tension. A group of six distinct colors will differentiate data well, but may look garish. Color palettes often shoulder even more responsibility, such as advancing a corporate identity. Everyone associates certain colors with their experience of the world. This is why it is so much easier to make intuitive color palettes for maps, as we saw in the blues and greens of my hometown.

  Pleasing colors blend and do not provide good visual discrimination.

  WILLIAM CLEVELAND, 1985

  As we age, our eyes change. Our eyeball lens hardens, making fine print harder to see without reading glasses. Early onset of conditions like cataracts or glaucoma can also make it harder to see. Consider how color contrasts across your visual. A thin grey font on a white background might look slick, but it might also slow down readers.

  To label (color as noun), to measure (color as quantity), to represent or imitate reality (color as representation), and to enliven or decorate (color as beauty).

  EDWARD TUFTE, 1990

  At minimum, an effective color palette does not turn the viewer away. Ideally, it helps attract and hold attention while also informing effectively. You must balance many demands across color pursuits. The technical best practices—not too many categorical colors, bucketed palettes, steady luminance, meaningful associations, and sensitivity to impairments—are all good guardrails against making color mistakes. Some programmed default color palettes clear these best-practice hurdles, but risk leaving too much value on the table. All defaults are uninteresting because they are overused. Defaults are often poorly linked to the semantics of the data because defaults are not designed for the information your data might convey.

  Whenever possible, make intuitive palettes. … It's not always possible, of course (what color is electrical charge, or income?) but a fair number of datasets lend themselves to particular colors.

  ROB SIMMON, 2013

  Your human consideration is needed to color data. Begin by referencing inspirational sources, including existing visualization and other graphic design. Then prototyping, demonstrating, and technical guardrails can help the palette be refined into something that is technically, semantically, and emotionally superb. Superior colors encourage readers to engage and help data soar.

  Recognize that it's hard and that it's going to take time and effort. Point that out to your stakeholders, schedule some time for it, don't just brush it off. One major reason why people are so bad at color in data visualization is because they don't budget any time for it.

  ELIJAH MEEKS, 2018

  Color is one of the most powerful channels we have. It can convey categories, quantities, and emotions. Color is also one of our most unwieldy tools. It requires more navigation of the encoding-to-decoding process than position or size. Remember, the same single data dimension can be mapped to multiple channels to reinforce the message. Do not choose between circle size or circle color to represent a quantity if both can be used in harmony.

  Together, position, size, and color give us the tools necessary to humanize data and see information. Next, we are going to take these tools and use them to probe data for better and better forms.

  CHAPTER

  8

  EXPLORE TO CREATE

  And this I believe: that the free, exploring mind of the individual human is the most valuable thing in the world.

  JOHN STEINBECK, 1952

  A bearded man sits down at a long table in the middle of a rowdy hall, orders wine, and unrolls the packet under his arm. People take notice of the drawing supplies now covering the table. They peer over his shoulder and ask questions as he studies the revelers around him. Their likeness is captured with quick strokes. We are in Rome. It is 1511 and the man is Michelangelo. He is studying the faces of real people in pursuit of improving his apostles for a new commission from Pope Julius II, the Sistine Chapel's ceiling.

  Data Sketches

  The artist makes quick, rough, unfinished drawings for many reasons. Sketches hone technical skills. They also help one imagine, invent, and discover. Postage stamp-sized doodles, cranked out by the dozen, give flight to the exploration of ideas as mind and pencil meld. Sketched lines can be altered, darkened, or abandoned in favor of starting anew.

  Live figure drawing classes often begin with the model executing many poses in succession, sometimes for only a few seconds each. These “gesture drawings” serve many purposes. They warm-up the artists as they connect the model to the page for the first time in that session. They distill the figure drawing to its essence as there is only time for simple curves and shapes. The dynamic energy of the pose is frozen on a static page.

  An idea transformed to the page escapes the impermanence of your own thinking. Outsourcing idea storage frees up your limited working memory for new thoughts. The sket
ch also makes an idea more real. You can point to it and say there it is. Something new has taken on its own identity. Once externalized, a sketch may serve as spark for more ideas. For artists, sketching can also help focus subject, position, and composition—before attempting canvas or marble. Like them, you have to figure out how to generate better ideas faster so you are not dragged down by the burden of full production.

  The germ of an idea may often come in a reverie as a purely cognitive act, the major work of creative design is done through a kind of dialogue with some rapid production medium.

  COLIN WARE, 2008

  Data sketches are little pictures that help reveal what the data has to tell you. They are not the only way to investigate data, but they are a unique way of investigating data. Data sketches do not have to be pretty or conclusive. We are not yet polishing for presentation. That can wait. The purpose of sketching is to encourage you to ask more questions of the data so that you can create more sketches that are even more revealing. That is why this creative dialog must extend through many iterations. No single picture shows all. Position, size, and color give us all the tools we need to begin sketching. Start with whatever chart type is easiest to generate. It is often a chart that aligns closely with the data's native structure. Peek at the profile of a single data column on a histogram. Scatter two-dimensional dots on a plane. Raise categorical bars on a graph. Connect nodes by their links. See where points land on the map. The first step is to produce a picture of some data—any data dimension(s), any picture. At the outset, what matters most importantly is that you create something you can see so you can proceed to the next step: React to the first sketch.