Graphical representation is a crucial part in statistical data analysis. It has been developed since 1980s. William S. Cleveland and Robert McGill are pioneers for the theory, and later it developed into different directions. The main goal for graphical perception is to extract quantitative information from graphs, and the perception is very important to determine the performance of graphs.

Cleveland and McGill (1984), through their study on Graphical Perception, set crucial ground work in the theory of graphic perception by identifying key human elementary perceptual tasks that are the most basic form of human perceptual activities when extracting quantitative information from graphics. They pointed out how the same amount of quantitative information represented in certain visual forms can be more effective in delivering the true message than others due to various psychological and cognitive reasons. They also proposed different methods for reducing visual bias and observation errors.

Jeffrey Heer and Michael Bostock write a paper about assessing the viability of crowdsourcing for evaluating the large design space of data visualization in 2010. They use Mechanical Turk created by Amazon as a platform to do graphical perception experiments. They firstly replicate previous graphical perception experiments and their crowdsourced results are a good match. Although the variation of the results are increased, the scalability of the platform can offset it. As crowdsourcing provide a cost-effective way to do experiments, it can be combined with laboratory experiments to cross check each other and leverage their own strengths. Besides, they also provide new aspects of results on rectangular area jugements, chart size and gridline spacing.

For example, Below are two chart legends taken from data visualizations done in Tableau.

In both of these examples, notice how the legend bar appears to be a different size at each end. The light ends appear to be wider than the dark ends. The illusion here is that the bar is getting narrower as our eyes move from the left side of the bar to the right side. They are actually true rectangles, but our minds are playing a trick on us. In this case, the gray border around the legend appears as a border on the left side, but on the right side, the border blends into the darker color of the legend which makes it look like it disappears.

This is a very subtle and minor example, but this is one of the reasons that data visualization experts avoid gradient effects and pay such close attention to the use of color. In this particular case, the sequential color scheme is used on a chart or map and this is the legend, so the gradient color isn’t used for an effect, but rather encoding a range of data, going from light for the low value to dark for the high value.

There are a few solutions to solving this particular issue. One easy method is to apply a darker border to the legend. In the example below, the black border moves the blending effect further to the right side, therefore reducing the effect of the illusion.

This helps quite a bit, but it doesn’t solve it completely. In the original example which is the default in Tableau, the light gray border disappears in the middle of the bar and blends into the bar color. Using a black border moves that blending to the right side of the legend.

In the next example, there is a dark border around a light border, and the visual perception problem is solved. Notice in this example there appears to be white space on the right side of the legend that blends in on the left side. The double border helps visualize an equal size legend as the eye moves from left to right.

Below is a visual comparison of the 3 options side by side:

In general, Graphical perception is extremely important to make data visualization efficient, and there is still a lot of work that needs to be done to figure out more efficient ways that allow for visualizations to be interpreted by faster parts of the brain that require less energy, which results in more efficient cognition.

References

[1] Cleveland, McGill - 1984 - Graphical Perception Theory, Experimentation, and Application to the Development of Graphical Methods

[2] Heer, Bostock - 2010 - Crowdsourcing Graphical Perception Using Mechanical Turk to Assess Visualization Design