CHAPTER 9
Cohesive Data Messages
So far we've covered vagueness and abstraction, data literacy, preparation, and sizing. Let's dig deeper into the visualizations themselves and figure out how visual encodings of data make sense to their intended audience. Cohesion is achieved when parts of a message bind together, making it understandable. Ideas that stick together can build a complete unit that readily stands on its own. People interact with their data as part of a flow that we call analytical conversation. Analytical conversations take place in a variety of ways. They happen during the analysis as we examine the data directly. They can be scripted for others to consume. Novel approaches to analytical conversations include incorporating natural language into the process, through innovations such as Tableau's Ask Data and Microsoft's Power BI, for example. The binding characteristic is that the response aligns to our request.
Cohesion means ideas work together to build a unified whole, which helps conversation interlink in purposeful ways, and the basic parts adhere to grammar. Like language, text and visual mediums also need to be coherent to ensure clarity. Coherence is a semantic property of conversation where content stays relevant and flows from one state to another (van Dijk, 1977).
As you saw in Chapter 6, a literate culture builds commonly shared approaches to expositions so that writers and readers know what to expect. Images paired with text also have cohesion by enhancing the piece without being redundant or in opposition to the content (Serafini, 2014). We will explore the interplay of text and charts more in the next chapter.
Comics also provide a valuable framework for cohesion. As a medium between art and story, comics balance text and images to drive a story. Comics and graphic novels rely on the ability to read space, illustrations, and stylized text for additional cues. They feature both literal interactions and ones created figuratively by use of proximity. Figure 9.1 gives a sense of how visual reading plays with placement. In this example, two cats face off between a water glass. Between the nine frames, the story itself can be read a few ways. We can follow a conventional Z-reading pattern, but we are also free to read the story in a downward direction as well.
FIGURE 9.1 Sample comic drawn by author
Comics use transitions between panels to support cohesion. Most comics favor transitions around changes in actions, subjects, scenes, or even aspects like camera angles to create unity in a narrative (McCloud, 1993). These mechanisms are intermixed. A few panes may set the scene by showing aspects first, then switching to subjects and action in a single page. If you were to look at a comic, you'd see empty space between the panels that contain the illustrations and the dialog. In the comics world, this space is known as the gutter. The gutter provides breathing space between conversation chunks to support coherence and for closure to happen. McCloud describes closure as “observing the parts, but perceiving the whole.” This idea is important because comics are a static medium where real-time actions don't happen. Rather, the author needs to convey the passage of time and movement through the comic. If you'd like to learn more about the various transition techniques that comic authors apply, it would be worthwhile to read McCloud's book. Visualizations also use techniques similar to comics to expose information and drive the interaction.
Cohesion in Designing Visualizations
Because they are scripted conversations, designing data visualizations for others to consume parallels comics and writing. Unlike a printed medium, digital charts can filter others, expand or contract, and animate to show changes. They can recolor data points, reorder sorts, and highlight various marks. As charts become a primary means for interpreting information through increased graphicacy skills, cohesion demands between them increase. This new medium provides novel ways to associate charts semantically.
We've seen this phenomenon play out with the rise of COVID-19 tracking dashboards. To communicate public health at scale, governments for countries, states, counties, and even cities have designed numerous tools to help the broader public understand various facets of the novel disease (Patino, 2021). The analytical dashboard is now mainstream.
Despite the new reach of data visualization, creating effective visualizations still remains a challenge for people designing the interactions. A scripted conversation requires effective prediction of the most relevant questions to ask the data. It demands building a composition that is well put together and guides users through the conversation without feeling unduly restrictive.
A semantic approach to visualization focuses on the interplay between charts, not just the selection of charts themselves. The approach unites the structural content of charts with the context and knowledge of those interacting with the composition. It avoids undue and excessive repetition by instead using referential devices, such as filtering or providing detail-on-demand. A cohesive analytical conversation also builds guardrails to keep users from derailing from the conversation or finding themselves lost without context. Functional aesthetics around color, sequence, style, use of space, alignment, framing, and other visual encodings can affect how users follow the script.
Color
Practitioners often find color to be one of the hardest attributes in a visualization. Color can make or break an otherwise stellar composition. Common advice encourages “getting it right in black and white” (Stone, 2006). Using monochrome design can help identify points of breakdown where color can clarify or identify the focus. We can see two versions of the same dashboard in Figure 9.2. The first cut is in gray. The first two charts at the top are legible without color. The bottom is where meaning starts to break down. With color, the message becomes clearer.
Semantic use of color supports the understanding of what the visualization is conveying. When color is used for a specific paradigm, those using the visualization can follow that paradigm. One paradigm might be using a specific color to highlight selections on an otherwise monochrome visualization. In others, color may be categorical but match associations with the time of day, such as in Kelly Martin's work shown in Figure 9.3.
Color can also help direct attention to differences in the data. In Figure 9.4, color highlights specific metrics, such as blue for sales, teal for profit, and gray for amounts. While never overtly stated, the dashboard subtly supports the consumer during the analysis.
Sequence
As we saw in comics, placement and design elements affect the interpretation of sequence. To create cohesion, charts should add or refine the idea in mind. Literate societies create common exposition styles. Practitioner and Tableau Hall of Fame Visionary Adam McCann (2019) identifies five types of dashboards commonly used in practice:
FIGURE 9.2 Identifying where color clarifies
FIGURE 9.3 Semantic color use in Kelly Martin's visualization
FIGURE 9.4 Semantic color use around metrics
- Key performance indicator (KPI), or a baseball card style with repeating assets for various metrics
- Question and answer (Q&A), a style driven by various questions that may not be cohesive as a piece
- Top down, a cohesive narrative that starts at a high level and becomes more granular
- Bottom up, a narrative that starts with the details and contextualizes them to a greater whole
- One big chart with nuance from filters
These different exposition styles are patterns for the order in which we expose information. As we enter into certain types of analytical conversations, we expect the conversations to flow in a predictable and cohesive manner. A KPI dashboard, for example, uses redundant structures across specific dimensions or measures to convey information. A dashboard with a top-down exposition style provides high-level information first and clarifies downward, while a bottom-up dashboard starts with the details and clarifies them against the larger picture. Other exposition styles also exist, and some may blend across some of these categories.
Figure 9.5 provides an example of a top-down dashboard that doesn't follow the expected narrative formula. While the map drives the interaction, some higher-level details are at the bottom of the reading order. The details are exposed first, but the higher level doesn't add context. It feels out of place, despite the interaction at the top.
FIGURE 9.5 A non-cohesive sequence
When we redirect the sequence of this example, we get a higher-level view first. Figure 9.6 corrects the errors in sequence. The map still drives the interaction, but the sequence also helps reference what was seen before. By moving the interaction lower, we first set the tone and then allow the interaction to further nuance the information.
FIGURE 9.6 A cohesive sequence
Visual Supports and Style
We cue users with visual supports and style. A style guide is a common practice for practitioners. It's the organization of elements such as titles, headers, and body text into a hierarchy of styles. These styles create unity and help the consumer navigate the visualization. Consistency improves learnability and establishes context. Style tiles provide a quick view of branding in a singular visual pane, like the one shown in Figure 9.7. They are one way of communicating a standardized look-and-feel. Style tiles can set the fonts, gridlines, colors, and assets. They are easy to wrap into an existing analysis so that other analysts can leverage them.
FIGURE 9.7 Simplified style tile
Do lines at zero have a particular style or color? Are gridlines brought forward or should they be nonexistent for chart types? These stylistic elements tell those interacting with the charts where to focus as they dig deeper into the interaction. Are charts treated as separate elements of meaning or, by drawing gridlines across to all, are they one cohesive unit like small multiples? In comics, the style includes what features are exaggerated versus what are realistic. With charts, we can exaggerate the emphasis on the data ink or draw the focus from the data ink to either a broader picture (several charts) or the data ink in context.
In addition to style tiles, framed layouts are another means of creating and applying style. Figure 9.8 shows two layouts. The first is the frame used for Figure 2.9 earlier in this book. The second is an outline view of how the dashboard in Figure 2.9 was used. While proportionally similar, the frame likely guides toward a different design process due to the design elements.
FIGURE 9.8 Frames differentiated by design elements
These frames are semantic. They help create a visual prioritization for reading, acting as syntax or a way of discerning reading order. The first frame encourages a down and over (N) reading style, while the second frame is more open to a Z-reading pattern. Multimodal reading allows the designer to encourage a reading style that best fits the message.
Like multimodal reading, data literacy relies on both primary literacy skills and numeracy skills to truly make sense of the third layer: reading and understanding graphs. Charts codify numbers visually into parameters, using stylized marks to embed additional layers of meaning and space to provide quantitative relationships. Beyond the individual chart, data visualizations create ensembles of charts. Various bento box layouts help guide the design process and cue the end user on how to read the chart, as we've discussed in earlier chapters.
Interactivity and text play a role as visual supports. We will further explore the interplay of text with charts in the next chapter. Interactivity allows a visualization to be explored as one layout visually. Interacting changes what we see, encouraging us to reexplore. Transitions can help provide context to how the data has shifted with the interaction.
Use of Space (Shape)
Charts create form through both positive and negative space use. The shapes of the data can draw attention to specific areas by creating either a positive or negative space. In comics, negative space can be used starkly to draw attention to the focal point. In Figure 9.9, a man stretches out his arm beyond the pane's view. There are two blank panes and then finally, there's the arm. Visually, it plays with the semantics of how we read comics and teases the viewer.
FIGURE 9.9 A comic using negative space to make a point
Charles Barsotti/CartoonStock Ltd
Positive and negative space help create balance, but they also draw interest. In Kelly's composition (modified in Figure 9.10), several shapes attract attention. We have the prominent use of the US map. Icons at the top clarify animals, while in-chart icons clarify direction of the plane. The scatterplot itself uses a distinct “crash” shape and creates a larger shape around the arced trend line. The use of space draws the eye through the visualization and creates a sense of movement.
FIGURE 9.10 Kelly Martin's viz modified to draw emphasis to use of space
Chart choices can also create weight within the entire composition. Presenting information as a comprehensive visualization, such as in a dashboard, requires thinking beyond individual charts. In writing, we not only craft sentences, but write the composition as an entire piece. Certain sentences may drive the writing more, but all sentences play a role in conveying the message. In Kelly's visualization, the map may be the largest piece, but the clincher is the scatterplot.
FIGURE 9.11 Effects of chart design on negative space
Crafting the broader narrative and shape may affect individual chart choices. In the visualization in Figure 9.11, parameters around alignment are used to convey relationships. To do this, the chart itself was changed from an unsynchronized axis to one that was synchronized to focus on the relationship in context to sales.
Use of space can help guide users to areas of interaction by making the interacted item larger. Shape may also be created by using other parameters to design layouts that weigh focuses differently.
Alignment
The alignment of charts builds cohesion by communicating relationships. Alignment paradigms may be based around outside borders, titles, and axes but also the data ink itself. Figure 9.12 shows Kelly Martin's exceptional craft around using alignment to build balance. Lines drawn around key areas of emphasis highlight the alignment paradigms at play in this work.
FIGURE 9.12 Kelly Martin's visual poetry
Aligning on data ink can be a powerful way to build relationships across charts. It can be used to obscure the lines between charts, making the composition feel more seamless. Kelly uses this effect to make the scatterplot not only a focal point but also the canvas on which other elements are added.
Alignment paradigms can also influence the layout design needed. In Figure 9.13, the shaded boxes are specifically designed to create ink and reinforce the relationship to the other items. The layout added to the alignment further supports this relationship.
FIGURE 9.13 Coral highlighting the alignment relationships
Disrupting alignment across points alters the proposed reading flow. Comics make use of this concept throughout the story in different patterns. Some reading patterns within comics can be open or even ambiguous by design. Note how the comic at the beginning of this chapter (Figure 9.1) uses alignment.
Figure 9.13 also shows how the charts outside the layout container don't align on data ink to the charts inside. This helps further reiterate that these are different charts. Among the two charts outside the container, they align on data ink and their titles are shifted accordingly.
Register
Register determines the degree of formality, comfort with the topic, and experience reading charts that the targeted audience has. As we design dashboards as scripted conversations with data, we look at how we choose charts differently based on audience or register. Box plots are a perfect example of how charts get adapted for register. Schools, statisticians, and performance analysts may use them often. Outside of certain environments, they read like jargon. We'll explore register more in Part C as we look at visual communication in Chapter 12.
Analytical Conversation
Functionally aesthetic visualization is most effective when users can focus on the process of interacting and exploring their data, without getting bogged down by the mechanics of doing so. When we converse with other people, we try to effectively convey the main idea, often using intonation, gesture, and emotion. Let's go back to our restaurant scene in Chapter 1 where we are dining in our cozy circle at a restaurant. A waiter approaches our table and the conversation goes like this.
| Waiter: | How are you folks today? |
| You: | Great, really excited to try the food. What's good here? |
| Waiter: | Oh, the seasonal sushi platter isn't to miss. It has seasonal vegetables with your choice of protein. A lot of people also enjoy our house Ramen Noodle Soup that you can see pictured there on the menu. |
| Your friend: | What about something that's hot? |
| Waiter: | Oh, we can adjust the spiciness on most of our menu items. If you're feeling really adventurous, you can try our Red Dragon, with tofu, avocado, jalapenos, cilantro on top with fresh, spicy albacore. |
Within the setting and context of this conversation, the waiter expects a brief tone-setting answer to decide how best to get an order. Questions like, “What's good here?” provide an opportunity for the waiter to pitch popular dishes, easy-to-miss delights, and seasonal fare. We've previously discussed the ambiguity of “hot.” Based on the context and semantics of what is being asked, the waiter alludes to the notion of spice seamlessly. The conversation should flow back and forth, perhaps with lulls while eating or drinking, or during bits of inattention to look at a phone or peruse the menu. Threads should bind the conversation, with new topics or themes clearly marked by natural transitions. These common threads create coherence. So how can we take some of these nuggets and apply them to creating effective analytical conversation?
We can draw inspiration from how people communicate when thinking about designing visualizations to support users conversing with data. Specifically, we explore how principles from pragmatics and cohesion can be applied to the flow of analytical conversations, specifically the notion of centering.
Conversational Centering
As we saw with the interaction between the diners and the waiter from the restaurant scene, conversations are more than just individual chunks of disparate dialog. We build off of what was said before, taking into consideration the context and relevance of what is currently being expressed. Conversational centering describes how the context of a conversation adjusts over time to maintain coherence, through transitional states that retain, shift, continue, or reset these conversational elements.
The process of conversing with data is most effective when users can focus on interacting and understanding the patterns presented and being in a state of flow. Pragmatics is particularly important for visual analysis flow, where questions often emerge from previous questions and insights. The principles of pragmatics are modeled based on the interaction behavior of a human-to-human conversation. Tory and Setlur (2019) adopted this conversational centering model to visualization flow, shown in Figure 9.14.
FIGURE 9.14 Conversational transitions model
A key insight of this model is that conversational transition states of continuing, retaining, and shifting apply to all parts of the visualization. Maintaining state in the visual encodings supports coherence, as abrupt changes to the visualization can be jarring and easily misinterpreted. Look at the earlier examples again and you can see effects of changing encodings. As a user interacts, they expect the results to follow a script that aligns with the original inquiry.
After interpreting a visualization (the thinking human in Figure 9.14), a user may continue their analytical conversation by formulating a new question. This analytical intent will ultimately drive a user's transitional goals (how they wish to transform the existing visualization to answer the new question), which in turn drive user actions. The research identified the following transitional goals:
- elaborate (add new data to the visualization)
- adjust / pivot (adapt aspects of the visualization)
- start new (create an altogether new visualization)
- retry (re-attempt a previous step that failed)
- undo (return to the prior state)
A key principle in visual analytics flow is the need to support interactive exploration and continue to build on the last question. A single static chart is rarely sufficient except in the simplest of investigative tasks. The user often needs to interact with their data, iteratively evolving both the questions and the visualization design. Direct manipulation is an effective interaction technique when one can easily point to the objects of interest. An icon may cue us to click. As we do, the data may shift with transitional movement helping us catch the changes. Chart marks might fade entirely or reorder. As we continue to interact, we may keep elements in mind: sales last week were around $200K, but this week they're only $120K. Elements within the chart, such as color and shape, may provide breadcrumbs of where we last clicked.
Natural language interfaces support analytical conversation, where a user can have a back-and-forth interaction with these tools. Evizeon (Hoque et al., 2018) is a system that supports pragmatics and coherence in analytical conversation. Figure 9.15 presents an example from that system.
The first query in Figure 9.15, “measles in the uk,” filters to measles cases in the United Kingdom. The user then types “show me the orange spike” and the system understands that the query is a reference to the line chart and annotates the spike. In the third query, the system interprets “mumps over there” as containing a reference to the United Kingdom and a different value in the disease attribute. It retains the filter on “United Kingdom” but updates disease from “measles” to “mumps.” “Epidemics here” is a reference to marks selected on the map with a mouse, so epidemic diseases in that selected region are highlighted.
Throughout this exchange, the user has been able to build on their prior queries and adapt the current state rather than starting over each time, just like how we speak with other people. Language is often incomplete or imprecise, relying on the audience to interpret using their contextual knowledge (i.e., speaker, topic, time, location, past dialog). These tendencies carry over into interactions with a visualization, where it is known that people use ambiguous language and may refer to items in the past. We will continue this conversation (pun intended) as we go into text and charts.
FIGURE 9.15 Example results of various forms of natural language interactions with a dashboard showing disease outbreaks in the world
Summary
Getting one chart to flow to the next smoothly relies on a variety of techniques that create cohesion. Comics also challenge us to think beyond standard patterns. While Z-reading can be an effective technique, framing can help bind units to encourage new paths. We can free our design experiences by playing with layout, drawing tighter or looser spaces, and aligning in ways that pivot the conversation in new directions. Analytical conversations are not linear, but they rely on logical transitions to shift directions. They build in ways to repair or lend nuance to a detail. Charts work together clarifying one another, faceting a new dimension, or directing attention to how a detail fits in the whole picture. Analytical conversations require cohesion to find insight.