Show and tell: approaches for effective figures

Le résumé fourni par la source

Humans are visual beings, and everything we look at initiates a cascade of cognitive reactions that hopefully ends in information transfer and understanding. Despite relatively well-understood cognitive pathways for visual messaging, scientific figures, charts, and other visuals are often developed based on familiar conventions that are disconnected from visual best practices. Fortunately, nearly all figures can be improved with a basic understanding of the characteristics of an effective figure along with the tools needed to create effective figures. “Trouble with you is the trouble with me, got two good eyes but we still don't see.” –Robert Hunter We have all experienced the magnetic pull of colorful, clear, and informative figures that tell a scientific story. But we have also looked at scientific figures that resulted in confusion or even elicited an almost visceral negative response. Why does such a disparity of experiences exist across data visuals? The answer lies in the ways in which we process visual information and whether the figure creator has used techniques that are compatible with this process. Much of our figure-making experience develops from what we like or dislike about figures we encounter, or we come across one of the many (helpful?) articles or graphics illustrating the ‘dos and don'ts’ of figure-making (Kelleher and Wagener 2011; Rougier et al. 2014; Greenacre 2016; Midway 2020). Yet rarely do we think about why we are drawn to certain figures or what hidden figure-making rules may have enabled its creation. Our responses to certain figures and graphical techniques are generated by known visual cognitive processes, and such processes have been studied for decades both in the context of scientific and non-scientific communication. This article provides suggestions for making effective scientific visuals in the context of understanding why these suggestions elicit a more positive response from the viewer and thus increase the author's ability to communicate the scientific message. We all use two cognitive processes to interpret visual information (Kahneman 2011; Padilla et al. 2018). The first is the fast, automatic, and impulsive visual system often referred to as System 1 decision making (or Type 1 processing; Evans and Stanovich 2013). System 1 is the reason why you can look at Fig. 1a and immediately interpret the relationship between the x and y variables. The second is System 2 decision making (or Type 2 processing) that requires logical, conscious interpretation. When examining Fig. 1b you are required to compare the individual relationships for birds and fish to the mean of the relationships, in order to come to a conclusion about the overall data. Anything that forces a viewer to engage in unnecessary conscious thought will reduce the effectiveness of a figure. Systems 1 and 2 operate simultaneously, and both need to be anticipated, yet the processes that initially engage a viewer are derived from System 1. For example, in Fig. 1a all extraneous information is absent, thus reducing the foreground effect in which the viewer may focus on only some visual information and come to an inaccurate conclusion (Stone et al. 1997). Furthermore, there is no clutter, such as a different numbers of decimal places on the x-axis or superficial colors. Visuals that minimize conscious thought, and therefore enhance System 1 decision making, are the reason for many of the “dos and don'ts” seen in publications relating to making scientific figures. But we cannot always tell our story with one regression line, and as such we must consider System 2 as our visuals take on more information. Yet we still want to reduce the amount of conscious thought required for System 2 thinking. One important tactic used to achieve this is to present the data in a manner that is congruent with the viewer's expectation. First described using cognitive fit theory (Vessey 1991; subsequent review Vessey 2006), the idea of viewer expectations is congruent with System 2 decision-making in which the required amount of working memory is increased if the information is not presented in the manner expected by the viewer (Padilla et al. 2018). For example, Fig. 1b requires less conscious thought than if the same information were presented as a table because we expect information to be presented as figures when we need to interpret relationships between variables. Conversely, we expect data to be presented in a table if we need to extract individual or specific numbers. To continue our examination of Fig. 1b, there are again no distracting errors that require unnecessary cognitive processing—the focus of the viewer is on interpreting the difference between the three lines presented. You are able interpret the figure without too much thinking because (1) the bird and fish data are labeled with different words, use different line formats, and are presented using colors that differentiate them from each other (and also from the mean), and (2) you can interpret the range of data compared to the average trend lines for bird and fish because both the individual data points and average lines are presented. The ease of figure interpretation is inversely proportional to the amount of conscious thought required to interpret it and working with Systems 1 and 2 will help reduce your viewer's cognitive load. In addition to the dual-process systems, it is important to understand the three levels of graphical cognition that you should expect viewers to use (Friel et al. 2001). First, a viewer will attempt to read the data, or otherwise orient themselves to what a bar height or point size or axis means. Second, a viewer will attempt to compare the data or read between the data, meaning they will look for comparisons, compositions, or other relations among the parts of the figure. Finally, the highest level of graphical cognition is reading beyond the data, where the viewer will make inferences or predictions beyond the data presented (reviewed in Galesic and Garcia-Retamero 2011; Okan et al. 2012). The process of graphical cognition takes place within both Systems 1 and 2, albeit much faster and more intuitively in System 1. Although these three levels of graphical cognition may seem simple, they remain a good reminder and way to audit the figures you make—is a viewer able to comprehend your figure on all three levels with minimal cognitive effort? Effective visuals are those that navigate the challenges of all three cognitive levels by reducing the amount of cognitive energy it takes to complete this process and arrive at an intended message. Recognizing if a viewer can comprehend a figure on all three levels of graphical cognition requires understanding how each level is affected by figure design. The first level of graphical cognition, reading the data, is affected by the design of axes, legends, captions, and the choice of figure used to represent the data (Fig. 2a; Boote 2014). If a viewer can't quickly and easily identify what the different figure components represent, comprehension and recall of the intended message will be hampered (Haroz and Whitney 2012; Borkin et al. 2015). The second level of graphical cognition, reading between the data, involves the recognition of relationships and interpolations between data (Boote 2014). This level of cognition is affected by the viewer's ability to interpret patterns. Pattern interpretation is made simpler by using commonly encountered figure types, making relationships easily identifiable (see suggestions below), and by linking the visualization to scientific concepts (Fig. 2b; Wang et al. 2012; Boote and Boote 2017). Finally, the third level of cognition, reading beyond the data, links the visualization to a hypothesis enabling interpretation of the current data and extrapolation of the message to other data (Boote 2014). The final level is mainly affected by emphasis on the message that you want the viewer to take away. In Fig. 2c,

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.