Title

Diverging color scheme

Headline

Display diverging data using two complementary color schemes that diverge from a common hue.

Illustration

Context

Many data distributions include a midpoint critical value where both ends of the data distribution are of interest. The midpoint critical value “…may be a mean, median, or zero value…” and the reader “…is often interested in patterns in the data that show clusters both above and below the critical value.” [Brewer, 1996, p. 79] Examples of diverging data distributions include anomalies and residuals.

Problem

Solution

Align perceptual orderings with logical orderings by displaying diverging data using a diverging (double-ended) color scheme. Craft color schemes using intensity to indicate magnitude and hue to indicate sign. In cases where the magnitude is relevant but the sign is not, consider using a more parsimonious grayscale intensity scheme. The combination of various pairs of single-hue sequential schemes with common endpoints produces many useful diverging schemes.

Forces

Implementation

Choosing an appropriate color scheme for diverging data need not be a difficult or complicated process. For some variables convention suggests the color scheme. Air temperature, for example, is often displayed using a diverging scale of red and blue with zero on the Celsius scale marking the transition. Cartographers have traditionally displayed surface elevation using a scheme of browns or greens and ocean depth using progressively darker blues. Brewer [1996] suggests the following hue combinations for diverging schemes: red/blue, orange/blue, orange/purple, yellow/purple, brown/blue, and yellow/blue.

Examples

Related Patterns

Consider using histogram to balance the data distribution with the color scheme (and also for display)

For data without a mid-point critical value, see sequential color scheme.

Title

Sequential color scheme

Headline

Display data that do not contain midpoint critical values using sequential color schemes

Illustration

Context

Many data distributions include a range of values without a significant midpoint. Absolute critical values may bound such distributions, as in the case of percentages, or the range of sampled data may arbitrarily define the endpoints, but there is no significant central value within the range of the data.

Problem

Spectral color schemes work poorly for sequential data, because spectral order produces no natural magnitude message in the viewer’s mind.

Solution

Display sequential data using a sequence of lightness steps combined with a single hue (e.g., Figure 1a) or with a hue transition.

Forces

Implementation

Beware that a hue transition may exhibit some of the characteristics of a diverging color scale. In particular, when yellow is employed as the hue, saturated yellow will tend to stand out from transitional hues that are darker.

Examples

Related Patterns

Title

Resolution indicates data quality

Headline

Displays with higher resolution than the underlying data can mislead the viewer

Illustration

Context

Array-oriented data typically represent phenomena measured, modeled, or recorded at a specific resolution, and often displayed as pixels or cells in a regular grid.

Problem

Interpolation and contouring algorithms smooth data by definition, but scale of measurement is an important characteristic of data and should not be obscured. High-resolution presentation of the data, or the use of high-resolution reference overlays, such as continental outlines or political boundaries, can obscure important limitations of the data or hide the nature of the underlying model. As an example, the viewer may make spurious inferences about climate change in the Rocky Mountains based on juxtaposition of a detailed United States map with the output from a climate model, which represents the Rocky Mountains only crudely.

Solution

Indicate the resolution of the underlying data by showing grid-cells and reference maps at their measured or recorded resolution. If effective resolution varies over the map, vary the display accordingly. Consider the array-oriented data as representing an array of pixels and display those pixels as faithfully as possible, without smoothing. If the array is insufficiently dense to provide a good visual representation without smoothing pixels, consider an alternate cartographic representation such as point symbols. The display should clearly show null or missing values using black, white, or an appropriate neutral color.

Forces

Implementation

Examples

Related Patterns

Title

Histogram-guided transformations

Headline

Viewing the data graphic on its own is often insufficient to understand the interaction of a color scheme with the data distribution.

Illustration

Context

Perceptually a color scheme is a selective transformation of the data.

Problem

Viewing the map alone is often insufficient for assessing the goodness of fit between the color scheme and the data, and for determining whether apparent patterns in the data are true and valid or merely display artifacts. Iteratively or interactively altering properties of the color scheme may also reveal patterns in the underlying data or model.

Solution

Use a histogram in combination with map display to evaluate color schemes. Viewing the two in combination may suggest appropriate mathematical transformations of the data for display.

combining the histogram with the data is also truth in advertising. Consider displaying the histogram along with the data if color is used selectively to accentuate certain values, features, or patterns.

Forces

Implementation

The histogram tool employed should be able to “ignore” missing values. “Standardized” data does not necessarily yield a normal distribution over any given map area.

Examples

Related Patterns

Title

Anomalies and residuals

Headline

Recognizing patterns from a single image of differences is often easier than comparing two images visually.

Illustration

Context

Exploring and highlighting patterns is a primary purpose of scientific data graphics. Measured in absolute terms, the variations that produce patterns are often small, but relative magnitude may be significant.

Problem

Depending on the data distribution, calculated differences between two data sets or variables (i.e., anomalies or residuals) may prove more useful than the two data sets themselves. Three commonly calculated differences are:

1.      Differences between two extremes, such as between July and January temperature.

2.      Differences between the data and some measure of central tendency, such as a mean or median value.

3.      Differences between measured data and data simulated by a model.

Solution

Mapping differences, in addition to or instead of absolute quantities often helps the viewer to recognize patterns, and allows the mapmaker to communicate more information in a smaller space.

Forces

Implementation

Examples

Related Patterns