Perception is defined as the ability to see, hear or become aware of something through our senses. It defines the way humans notice the world and their environment and what knowledge or information we gain from it. Perception is therefore present in every aspect of our life, including professional and educational. Throughout our life, we are often exposed to various types of stimuli from which we are expected to extract information. One of those stimuli is visual representation, more specifically graphics representing data. Data graphics are ever-present in today’s society, even more so in our age which is so much data-driven, and we are expected to make sense of them, and make decisions based on them. This requires us to accurately decipher the information presented to us in data visualization. In this paper, we will go over the role of perception in data visualization. Then we will explore different principles and rules that make perception more accurate on data visualizations objects and the resulting recommendations for representing data based on those. Finally we will discuss its significance in today’s information age.

Human perception supports the cognitive associated process. Cognition is defined as the way our brain learns or acquires knowledge through thoughts, experiences or senses. Therefore, the way our brain learns is directly connected to perception. And the way we extract information or knowledge from data visualization is no exception. Indeed, graphics perception helps us extract and define the relationship between the data. It is by mentally evaluating patterns like length, sizes, angles etc, that we can extract information from graphs. For example in this chart below, we can say with fairly strong confidence that NY’s population is smaller than Texas without any axis or number to guide us.

Alexandre and Talvares go into further detail to explain what principles our brain follows in order to make sense of the things we see. They discussed the well-known Gestalt Theory according to which the human brain configures information through the sensorial canals, perception and/or memory. In the 2010 paper, they describe the principles under which perceptive activity happens. As the authors say: “The Gestalt theory says that the perceptive activity is subordinated to a basic factor of Prägnanz (good shape). An object is Prägnanz if it expresses any characteristic in a sufficiently strong way to be obvious, to be imposed and to be easily evocative. The Prägnanz characteristic images are considered in the laws of the Gestalt Theory, which explain the structural and functional principles of the perceptive field. These laws establish the shape as the constituent elements of an image that can be perceived.” (2010). The principles they mentioned are the following:

− Proximity: occurs when elements are placed together in space and time. They tend to be perceived as a group, even when they are not similar.

− Similarity: elements that have similar or equal characteristics tend to be grouped in sets; the similarity occurs mainly in terms of color, shape and texture. Normally, similarity is not overlapped by proximity.

− Closing: elements are disposed in a certain way in order to form an almost closed outline or incomplete shape, which could become regular and stable. Thus, they are able to become a unity; this refers the tendency of human perception to realize complete shapes. Humans perceive the whole by filling the missing data.

− Simplicity: elements are perceived more easily when they present symmetry, regularity and are without textures.

− Continuity: the human perception has a tendency to orientate the elements that look like appear to build a pattern or a flow along a common direction; so, continuity of direction and continuous ligaments between elements are easier to of be perceived than the ones that present abrupt modifications in their direction.

− Figure/Background: any perceptive field can be divided into a figure and on a background. The figure is distinguished from the background by characteristics like: size, shape, color and position. The object is only perceived as a figure after being separated from the background.

These principles are critical to the design and creation of data visualization objects. They are what allow our brain to make sense of data when graphically displayed in a certain manner such as scatterplots of two different data sets or images scattered over a blank plane as opposed to arranged in a familiar shape.

Now that we have covered the basics of the role of perception in data visualization, it is important to explore the ways in which we can maximize its impact in our ability to accurately get a sense of the data presented to us. Several studies have been conducted in this aspect. One of the most significant is the “position-length” and “position-angle” experiments conducted by Cleveland and McGill in 1984. Their goal with these experiments was to construct a graph that uses the highest ranked elementary perceptual tasks. The experiments are described and the findings discussed in their paper titled “Graphical Perception: Theory, Experimentation, and Application to the Development of Graphical Methods”. In this article, the authors identify the essential elements of perception then conduct the mentioned experiments in order to determine which elements are more accurate. Here is the list of the 10 elementary perceptual tasks, ordered from most accurate to least accurate which was confirmed by their experiment: position common scale and position non-aligned scale; length, direction, and angle; area; volume and curvature; and finally shading, and color saturation. From these findings, key takeaways were made. One of them is the idea that for data analysis, triple scatterplots is a great way to represent 3-dimensional data. In addition to that, the authors recommend showing differences by a single line chart, which would display the actual difference as opposed to two lines showing the absolutes. Another one is the idea that bar charts should be used in replacements of pie charts due to the low rank of the angle perceptual task. On the topic of pie charts, there has been lots of discussion about the validity of pie and donuts charts as a meaningful way to convey information that could accurately be interpreted. This is partly due to Cleveland and McGill’s findings of the angle being evaluated at a pretty low accuracy level. Drew Skau and Robert Kosara decide to focus their study on the pie and donuts charts. In their 2016 article titled “Arcs, Angles, or Areas: Individual Data Encodings in Pie and Donut Charts. Computer Graphics Forum”, they break down the elements used by users to make sense of the data displayed in pie and donuts charts and try to see which one do people use the most and what is each one’s influence into the overall accuracy of a pie or donut chart. According to them “the data is encoded into pie and donut charts 3 ways: through arc length, center angle and segment area”. What they found is that area and arc length charts generated more accurate results than angle charts. This was not expected, as the common idea was that the angle was the most critical part of such charts. Their experiment allowed them to establish a ranking of the different types of pie and donut charts presented in the study. It goes as follow: baseline donut and baseline pie → arc and area → angle pie→ angle donut.

With the exponential growth of data available and the need to make sense of it all for businesses to gain competitive advantage, it has become paramount to be able to produce graphics that decision makers can accurately read and extract information from. We have seen a proliferation of data visualization tools and features in all the BI platforms and programming tools. Graphics Perception needs to be taken into account and incorporated in the creation and development of those data visualization features because we want to maximize the readability of the charts and graphs presented to the intended audience. Indeed, most of these tools have a “graphics suggestion” feature where based on the data presented by the user, the program is able to recommend the appropriate graph and subsequently create such graphic. It is therefore important for the developers to be informed on the “readability accuracy” of the different perceptual tasks in order to know which charts to setup as the “highly recommended option” for a specific data type. In that sense, graphics perception plays a very important role on data visualization. It is therefore necessary to continue studies and experiments to gain further insights into what else we can find or do to make it work in our favor. One significant obstacle is the cost and resources needed to conduct traditional surveys and experiments. To address this issue, Heer and Bostock provide us with an alternative: “crowdsourcing”. Crowdsourcing is the concept in which web workers complete one or more small tasks, often for micro-payments on the order of $0.01 to $0.10 per task. The authors’ goal was to see if crowdsourcing was a valid alternative to the traditional way of conducting such experiments. They have proven that this way of conducting the study reduces costs and recruiting time, and could be scalable as long as the conductors remain equitable towards their web workers by compensating them fairly.

As a summary, we can say that perception plays a key role in our ability to extract information from data visualization. The studies mentioned earlier provide us with useful insights into the principle that affect our perception of charts and the most efficient tools in order to maximize accuracy. These recommendations need to be taken into account by data visualization tools developers in order to provide us with the best graphics to convey information about the data depending on its type.

Sources

Heer, Jeffrey & Bostock, Michael. (2010). Crowdsourcing graphical perception: Using Mechanical Turk to assess visualization design. Conference on Human Factors in Computing Systems - Proceedings. 1. 203-212. 10.1145/1753326.1753357. https://www.researchgate.net/publication/221515971_Crowdsourcing_graphical_perception_Using_Mechanical_Turk_to_assess_visualization_design

S. Cleveland, William & Mcgill, Ron. (1984). Graphical Perception: Theory, Experimentation, and Application to the Development of Graphical Methods. Journal of the American Statistical Association. 79. 531-554. 10.1080/01621459.1984.10478080. https://www.researchgate.net/publication/229099907_Graphical_Perception_Theory_Experimentation_and_Application_to_the_Development_of_Graphical_Methods

Skau, Drew & Kosara, Robert. (2016). Arcs, Angles, or Areas: Individual Data Encodings in Pie and Donut Charts. Computer Graphics Forum. 35. 121-130. 10.1111/cgf.12888. https://www.researchgate.net/publication/304811503_Arcs_Angles_or_Areas_Individual_Data_Encodings_in_Pie_and_Donut_Charts

Sternadt Alexandre, Dulclerci & Tavares, Joao. (2010). Introduction of Human Perception in Visualization. International Journal of Imaging. 4. ., https://www.researchgate.net/publication/277068044_Introduction_of_Human_Perception_in_Visualization