Analyzing data : FORENSIC DESIGN ASSESSMENTS
saiptk
DataVisualisation_AHandbookforDataDrivenDesign-219-518.pdf
Part C Developing Your Design Solution
The Production Cycle
Within the four stages of the design workflow there are two distinct parts. The first three stages, as presented in Part B of this book, were described as ‘The Hidden Thinking’ stages, as they are concerned with undertaking the crucial behind-the-scenes preparatory work. You may have completed them in terms of working through the book’s contents, but in visualisation projects they will continue to command your attention, even if that is reduced to a background concern.
You have now reached the second distinct part of the workflow which involves developing your design solution. This stage follows a production cycle, commencing with rationalising design ideas and moving through to the development of a final solution.
The term cycle is appropriate to describe this stage as there are many loops of iteration as you evolve rapidly between conceptual, practical and technical thinking. The inevitability of this iterative cycle is, in large part, again due to the nature of this pursuit being more about optimisation rather than an expectation of achieving that elusive notion of perfection. Trade- offs, compromises, and restrictions are omnipresent as you juggle ambition and necessary pragmatism.
How you undertake this stage will differ considerably depending on the nature of your task. The creation of a relatively simple, single chart to be slotted into a report probably will not require the same rigour of a formal production cycle that the development of a vast interactive visualisation to be used by the public would demand. This is merely an outline of the most you will need to do – you should edit, adapt and participate the steps to fit with your context.
There are several discrete steps involved in this production cycle:
Conceiving ideas across the five layers of visualisation design. Wireframing and storyboarding designs. Developing prototypes or mock-up versions.
219
Testing. Refining and completing. Launching the solution.
Naturally, the specific approach for developing your design solution (from prototyping through to launching) will vary hugely, depending particularly on your skills and resources: it might be an Excel chart, or a Tableau dashboard, an infographic created using Adobe Illustrator, or a web-based interactive built with the D3.js library. As I have explained in the book’s introduction, I’m not going to attempt to cover the myriad ways of implementing a solution; that would be impossible to achieve as each task and tool would require different instructions.
For the scope of this book, I am focusing on taking you through the first two steps of this cycle – conceiving ideas and wireframing/storyboarding. There are parallels here with the distinctions between architecture (design) and engineering (execution) – I’m effectively chaperoning you through to the conclusion of your design thinking.
To fulfil this, Part C presents a detailed breakdown of the many design options you will face when conceiving your visualisation design and provides you with an appreciation of the key factors that will influence the actual choices you make. The next few chapters are therefore concerned with the design thinking involved with each of these five layers of the visualisation design anatomy, namely:
Chapter 6: Data representation Chapter 7: Interactivity Chapter 8: Annotation Chapter 9: Colour Chapter 10: Composition
The sequencing of these layers is deliberate, based on the need to prioritise your attention: what will be included and how will it appear. Initially, you will need to make decisions about what choices to make around data representation (charts), interactivity and annotation. These are the layers that result in visible design content or features being included in your work. You will then complete your design thinking by making decisions about the appearance of these visible components, considering their colour and composition.
220
Conceiving: This will cover all your initial thinking across the various layers of design covered in the next few chapters. The focus here is on conceiving ideas based on the design options that seem to fit best with the preparatory thinking that has gone before during the first three stages. As you fine-tune your emerging design choices the benefit of sketching re- emerges, helping you articulate your thoughts into a rough visual form. As mentioned in Chapter 3, for some people the best approach involves sketching with the pen, for others it is best expressed through the medium of technical fluency. Whichever approach suits you best, it is helpful to start to translate your conceptual thinking into visual thinking, particularly when collaborating. This sketching might build on your instinctive sketched concepts from stage 1, but you should now be far better informed about the realities of your challenge to determine what is relevant and feasible.
‘I tend to keep referring back to the original brief (even if it’s a brief I’ve made myself) to keep checking that the concepts I’m creating tick all the right boxes. Or sometimes I get excited about an idea but if I talk about it to friends and it’s hard to describe effectively then I know that the concept isn’t clear enough. Sometimes just sleeping on it is all it takes to separate the good from the bad! Having an established workflow is important to me, as it helps me cover all the bases of a project, and feel confident that my concept has a sound logic.’ Stefanie Posavec, Information Designer
Wireframing and storyboarding: Wireframing involves creating a low- fidelity illustration of the potential layout for those solutions that will generally occupy a single page of space, such as a simple interactive visualisation or an infographic. There is no need to be too precise just yet, you are simply mapping out what will be on your page/screen (charts, annotations), how they will be arranged and what things (interactive functions) it will do. If your project is going to require a deeper architecture, like a complex interactive, or will comprise sequenced views, like presentations, reports or animated graphics, each individual wireframe view will be weaved together using a technique called storyboarding. This maps out the relationships between all the views of your content to form an overall visual structure. Sometimes you might approach things the other way round, beginning with a high-level storyboard to provide a skeleton structure within which you can then form your more detailed thinking about the specific wireframe layouts within each page or view.
221
Prototypes/mock-ups: Whereas wireframing and storyboarding are characterised by the creation of low-fi ‘blueprints’, the development of mock-ups (for example, Figure C.1) or prototypes (the terms tend to be used interchangeably) involves advancing your decisions about the content and appearance of your proposed solution. This effectively leads to the development of a first working version that offers a reasonably close representation of what the finished product might look like.
Figure C.1 Mockup designs for ‘Poppy Field’
Testing: Once you have an established prototype version, you must then seek to have it tested. Firstly, you do this ‘internally’ (i.e. by you or by collaborators/colleagues) to help iron out any obvious immediate problems. In software development parlance, this would be generally consistent with alpha testing. Naturally, beta follows alpha and this is where you will seek others to test it, evaluate it, and feedback on it. This happens regardless of the output format; it doesn’t need to be a digital, interactive project to merit being tested. There will naturally be many different aspects to your proposed solution that will need checking and evaluating. The three principles of good visualisation design that I presented earlier offer a sensible high-level structure to guide this testing:
Trustworthy design testing concerns assessing the reliability of the work, in terms of the integrity of its content and performance. Are there any inaccuracies, mistakes or even deceptions? Are there any design choices that could lead to misunderstandings? Any aspects in how the data has been calculated or counted that could undermine
222
trust? If it is a digital solution, what is the speed of loading and are there any technical bugs or errors? Is it suitably responsive and adaptable in its use across different platforms? Try out various user scenarios: multiple and concurrent users, real-time data, all data vs sample data, etc. Ask the people testing your solution to try to break it so you can find and resolve any problems now. Accessible design testing relates to how intuitive or sufficiently well explained the work is. Do they understand how to read it and what all the encodings mean? Is the viewer provided with a sufficient level of assistance that would be required as per the characteristics of the intended audience? Can testers find the answers to the questions you intended them to find and quickly enough? Can they find answers to the questions they think are most relevant? Elegant design testing relates to questions such as: Is the solution suitably appealing in design? Are there any features which are redundant or superfluous design choices that are impeding the process of using the solution?
Who you invite to test your work will vary considerably from one project to the next but generally you will have different possible people to consider participating in this task:
Stakeholders: the ultimate customers/clients/colleagues who have commissioned the work may need to be included in this stage, if not for full testing then at least to engage them in receiving initial concept feedback. Recipients: you might choose a small sample of your target audience and invite those viewers to take part in initial beta testing. Critical friends: peers/team/colleagues with suitable knowledge and appreciation about the design process may offer a more sophisticated capacity to test out your work. You: sometimes (often) it may ultimately be down to you to undertake the testing, through either lack of access to other people or most typically a simple lack of time. To accomplish this effectively you have to find a way almost to detach yourself from the mindset of the creator and occupy that of the viewer: you need to see the wood and the trees.
The timing of when to seek feedback through testing/evaluation will vary across different contexts again. Sometimes the pressure from stakeholders
223
who request to see progress will determine this. Otherwise, you will need to judge carefully the right moment to do so. You don’t want to get feedback when it is too late to change or you have invested too much effort creating a prototype that might require widespread changes in approach. Likewise, it can be risky showing far-too-undercooked concepts to stakeholders or testers when they might not have the capacity to realise this is just an early indication of the direction of travel. The least valuable form of testing feedback is when pedantic stakeholders spend time pointing out minutiae that of course need correcting but have no significance at this stage. No-one comes away with anything of value from this kind of situation.
‘We can kid ourselves that we are successful in what we “want” to achieve, but ultimately an external and critical audience is essential. Feedback comes in many forms; I seek it, listen to it, sniff it, touch it, taste it and respond.’ Kate McLean, Smellscape Mapper and Senior Lecturer Graphic Design
Refining and completing: Based on the outcome of your testing process, this will likely trigger a need to revisit some of the issues that have emerged and resolve them satisfactorily. Editing your work involves:
correcting issues; stripping away the superfluous content; checking and enhancing preserved content; adding extra degrees of sophistication to every layer of your design; improving the consistency and cohesion of your choices; double-checking the accuracy of every component.
As your work heads towards a state of completion your mindset will need to shift from a micro-level checking back to a macro-level assessment of whether you have truly delivered against the contextual requirements and purpose of your project.
In any creative process a visualiser is faced with having to declare work as being complete. Judging this can be quite a tough call to make in many projects. As I have discussed plenty of times, your sense of ‘finished’ often needs to be based on when you have reached the status of good enough. While the presence of a looming deadline (and at times increasingly agitated stakeholders) will sharpen the focus, often it comes down to a fingertip sense of when you feel you are entering the period of diminishing
224
returns, when the refinements you make no longer add sufficient value for the amount of effort you invest in making them.
‘You know you’ve achieved perfection in design, not when you have nothing more to add, but when you have nothing more to take away.’ Antoine de Saint-Exupéry, Writer, Poet, Aristocrat, Journalist, and Pioneering Aviator
‘Admit that nothing you create on a deadline will be perfect. However, it should never be wrong. I try to work by a motto my editor likes to say: No Heroics. Your code may not be beautiful, but if it works, it’s good enough. A visualisation may not have every feature you could possibly want, but if it gets the message across and is useful to people, it’s good enough. Being “good enough” is not an insult in journalism – it’s a necessity.’ Lena Groeger, Science Journalist, Designer and Developer at ProPublica
‘It was intimidating to release to the public a self-initiated project on such a delicate subject considering some limitation with content and data source. But I came to appreciate that it’s OK to offer a relevant way of looking at the subject, rather than provide a beginning-to-end conclusion.’ Valentina D’efilippo, Information Designer, discussing her ‘Poppy Field’ project that looked at the history of world conflicts and the resulting loss of life
Launching: The nature of launching work will again vary significantly based, as always, on the context of your challenge. It may simply be emailing a chart to a colleague or you might be presenting your work to an audience. For other cases it could be a graphic going to print for a newspaper or involve an anxious go-live moment with the launch of a digital project on a website, to much fanfare and public anticipation. Whatever the context of your ‘launch’ stage, there are a few characteristic matters to bear in mind – these will not be relevant to all situations but over time you might need to consider their implications for your setting:
Are you ready? Regardless of the scope of your work, as soon as you declare work completed and published you are at the mercy of your decisions. You are no longer in control of how people will interpret your work and in what way they will truly use it. If you have particularly large, diverse and potentially emotive subject matter, you will need to be ready for the questions and scrutiny that might head in your direction.
225
Communicating your work is a big deal. The need to publicise and sell its benefits is of particular relevance if you have a public-facing project (you might promote it strongly or leave it as a slow burner that spreads through ‘word of mouth’). For more modest and personal audiences you might need to consider directly presenting your work to these groups, coaching them through what it offers. This is particularly necessary on those occasions when you may be using a less than familiar representation approach. What ongoing commitment exists to support the work? This clearly refers to specific digital projects. Do you have to maintain a live data feed? Will it need to sustain operations with variable concurrent visitors? What happens if it goes viral – have you got the necessary infrastructure? Have you got the luxury of ongoing access to the skill sets required to keep this project alive and thriving? Will you need to revise, update and rerelease the project? As I discussed in the contextual circumstances, will you need to replicate this work on a repeated basis? What can you do to make the reproduction as seamless as possible? What is the work’s likely shelf life? Does it have a point of expiry after which it could be archived or even killed? How might you digitally preserve it beyond its useful lifespan?
226
6 Data Representation
In this chapter you will explore in detail the first, and arguably the most significant, layer of the visualisation design anatomy: data representation. This is concerned with deciding in what visual form you wish to show your data.
To really get under the skin of data representation, we are going to look at it from both theoretical and pragmatic perspectives. You will start by learning about the building blocks of visual encoding, the real essence of this discipline and something that underpins all data representation thinking. Whereas visual encoding is perhaps seen as the purist ‘bottom- up’ viewpoint, the ‘top-down’ perspective possibly offers more pragmatic value by framing your data representation thinking around the notion of chart types. For most people facing up to this stage of data representation, this is conceptually the more practical entry point from which to shape their decisions.
To substantiate your understanding of this design layer you will take a tour through a gallery of 49 different chart type options, reflecting the many common and useful techniques being used to portray data visually in the field today. This gallery will then by supplemented by an overview of the key influencing factors that will inform and determine the choices you make.
6.1 Introducing Visual Encoding As introduced in the opening chapter, data representation is the act of giving visual form to your data. As viewers, when we are perceiving a visual display of data we are decoding the various shapes, sizes, positions and colours to form an understanding of the quantitative and categorical values represented. As visualisers, we are doing the reverse through visual encoding, assigning visual properties to data values. Visual encoding forms the basis of any chart or map-based data representation, along with the components of chart apparatus that help complete the chart display.
There are many different ways of encoding data but these always comprise
227
combinations of two different properties, namely marks and attributes. Marks are visible features like dots, lines and areas. An individual mark can represent a record or instance of data (e.g. your phone bill for a given month). A mark can also represent an aggregation of records or instances (e.g. a summation of individual phone charges to produce the bill for a given month). A set of marks would therefore represent a set of records or instances (e.g. the 12 monthly phone bills for 2015).
Attributes are variations applied to the appearance of marks, such as the size, position, or colour. They are used to represent the values held by different quantitative or categorical variables against each record or instance (or, indeed, each aggregation). If you had 12 marks, one for each phone bill during 2015, you could use the size attribute of each mark to represent the various phone bill totals.
Figure 6.1 offers a more visual illustration. In the dataset there are six records, one for each record listed. ‘Gender’ is a categorical variable and ‘Years Since First Movie’ is a quantitative variable. ‘Male’ and ‘43’ are the specific values of these variables associated with Harrison Ford. In the associated chart, each actor from the table is represented by the mark of a line (or bar). This represents their record or instance in the table. Harrison Ford’s bar is proportionally sized in scale to represent the 43 years since his first movie and is coloured purple to distinguish his gender as ‘Male’. Each of the five other actors similarly has a bar sized according to the years since their first movie and coloured according to their gender.
Figure 6.1 Illustration of Visual Encoding
The objective of visual encoding is to find the right blend of marks and attributes that most effectively will portray the angle of analysis you wish to show your viewers. The factors that shape your choice and define the notion of what is considered ‘effective’ are multiple and varied in their influence. Before getting on to there, let’s take a closer look at the range of
228
different marks and attributes that are commonly found in the data representation toolkit.
It is worth noting upfront that while the organisation of the ‘attributes’, in particular, suggests a primary role, several can be deployed to encode both categorical (nominal, ordinal) variables and quantitative variables. Furthermore, as you see in the bar chart in Figure 6.1, combinations of several attributes are often applied to marks (such as colour and size) to encode multiple values.
Although beyond the scope of this book, there are techniques being developed in the field exploring the use of non-visual senses to portray data, using variations in properties for auditory (sound), haptic (touch), gustatory (taste) and olfactory (smell) senses.
Figure 6.2 List of Mark Encodings
Figure 6.3 List of Attribute Encodings
229
230
Grasping the basics of visual encoding and its role in data visualisation is one of the fundamental pillars of understanding this discipline. However, when it comes to the reality of considering your data representation options you do not necessarily need to always approach things from this somewhat bottom-up perspective. For most people’s needs when creating a data visualisation it is more pragmatic (and perhaps more comprehensible) to think about data representation from a top-down perspective in the shape of chart types.
If marks and attributes are the ingredients, a chart ‘type’ is the recipe offering a predefined template for displaying data. Different chart types offer different ways of representing data, each one comprising unique combinations of marks and attributes onto which specific types of data can be mapped.
Recall that I am using chart type as the all-encompassing term, though this is merely a convenient singular label to cover any variation of map, graph, plot and diagram based around the representation of data.
Let’s work through a few examples to illustrate the relationship between some selected chart types demonstrating different combinations of marks and attributes.
To begin with Figure 6.4, visualises the recent fortunes of the world’s billionaires. The display shows the relative ranking of each profiled billionaire in the rich list, grouping them by the different sectors of industry in which they have developed their wealth. This data is encoded using the point mark and two attributes of position. The point in this deployment is depicted using small caricature face drawings representative of each individual – effectively unique symbols to represent the distinct ‘category’ of each different billionaire. Note that these are points, as distinct from area marks, because their size is constant and insignificant in terms of any quantitative implication. The position in the allocated column signifies the industry the individuals are associated with, while the vertical position signifies the rank (higher position = higher rank towards number 1).
For reference, this is considered a derivative of the univariate scatter
231
plot, which usually shows the dispersal of a range of absolute values rather than rank.
Figure 6.4 Bloomberg Billionaires
As seen in Chapter 1, the clustered bar chart in Figure 6.5 displays a series of line marks (normally described as bars). There are 11 pairs of bars, one for each of the football seasons included in the aggregated analysis. The attribute of colour is used to distinguish the bars between the two quantitative measures displayed: blue is for ‘games’, purple is for ‘goals’. The size dimension of ‘height’ (the widths are constant) along the y-axis scale then represents the quantitative values associated with each season and each measure.
Figure 6.6 is called a bubble chart and displays a series of geometric area marks to represent the top 100 blog posts on my website based on their popularity over the previous 100 days. Each circle represents an individual post and is sized to show the quantitative value of ‘total visits’ and then coloured according to the seven different post categories I use to organise my content.
Figure 6.5 Lionel Messi: Games and Goals for FC Barcelona
232
Figure 6.6 Image from the home page of visualisingdata.com
233
Figure 6.7 How the Insane Amount of Rain in Texas Could Turn Rhode Island Into a Lake
234
Figure 6.7 demonstrates the use of the form, which is more rarely used. My advice is that it should remain that way as it is hard for us to judge scales of volume in 2D displays. However, it can be of merit when values are extremely diverse in size as in this good example. The chart displayed contextualises the amount of water that had flowed into Texas reservoirs in the 30 days up to 27 May 2015. The size (volume) of a cube is used to display the amount of rain, with 8000 small cubes representing 1000 acre- feet of water (43,560,000 cubic feet or 1233.5 mega litres) to create the whole (8 million acre-feet), which is then compared against the heights of
235
the Statue of Liberty and what was then the world’s tallest building, the Burj Khalifa, to orient in height terms at least.
6.2 Chart Types For many people, creating a visualisation involves using tools that offer chart menus: you might select a chart type and then ‘map’ the records and variables of data against the marks and attributes offered by that particular chart type. Different tools will offer the opportunity to work with a different range of chart types, some with more than others.
As you develop your capabilities in data visualisation and become more ‘expressive’ – trying out unique combinations of marks and attributes – your approach might lean more towards thinking about representation from a bottom-up perspective, considering the visual encodings you wish to deploy and arriving at a particular chart type as the destination rather than an origin. This will be especially likely if you develop or possess a talent for creating visualisations through programming languages.
As the field has matured over the years, and a greater number of practitioners have been experimenting with different recipes of marks and attributes, there is now a broad range of established chart types. Once again I hesitate to use the universal label of chart type (some mapping techniques are not chart types per se) but it will suffice. While all of us are likely to be familiar with the ‘classic three’ – namely, the bar, pie and line chart – there are many other chart type options to consider.
To acquaint you with a broader repertoire of charting options, over the coming pages I present you with a gallery. This offers a curated collection of some of the common and useful chart types being used across the field today. This gallery aims to provide you with a valuable reference that will directly assist your judgements, helping you to pick (conceptually, at least) from a menu of options.
I have attempted to assign each chart to one of five main families based on their primary analytical purpose. What type of angle of analysis does each one principally show? Using the five-letter mnemonic CHRTS this should provide a useful taxonomy for organising your thinking about which chart or charts to use for your data representation needs.
236
I know what you’re thinking: ‘well that’s a suspiciously convenient acronym’! Honestly, if it was as intentional as that I would have tried harder to somehow crowbar in an ‘A’ family. OK, I did spend a lot of time, but I couldn’t find it and it’s now my life’s ambition to do so. Only then will my time on this planet have been truly worthwhile. In the meantime, CHRTS is close enough. Besides, vowels are hugely overrated.
Each chart type presented is accompanied by an array of supporting details that will help you fully acquaint yourself with the role and characteristics of each option.
A few further comments about what this gallery provides:
The primary name used to label each chart type as well as some further alternative names that are often used An indication of which CHRTS family each chart belongs to, based on their specific primary role, as well as a sub-family definition for further classification An indicator for each chart type to show which ones I consider to be most useful for undertaking Exploratory Data Analysis (the black magnifying glass symbol) An indicator for whether I believe a chart would typically require interactive features to offer optimum usability (the black cursor symbol) A description of the chart’s representation: what it shows and what encodings (marks, attributes) it is comprised of A working example of the chart type in use with a description of what it specifically shows A ‘how to read’ guide, advising on the most effective and efficient approach to making sense of each chart type and what features to look out for Presentation tips offering guidance on some of the specific choices to be considered around interactivity, annotation, colour or composition
237
design ‘Variations and alternatives’ offer further derivatives and chart ‘siblings’ to consider for different purposes
Exclusions: It is by no means an exhaustive list: the vast permutations of different marks and attributes prevents any finite limit to how one might portray data visually. I have, however, consciously excluded some chart types from the gallery mainly because they were not different enough from other charts that have been profiled in detail. I have mentioned charts that represent legitimate derivatives of other charts where necessary but simply did not deem it worthy to assign a whole page to profile them separately. The Voronoi treemap, for example, is really just a circular treemap that uses different algorithms to arrange its constituent pieces. While the construction task is different, its usage is not. The waterfall chart is a single stacked bar chart broken down into sequenced stages. Inclusions: I have wrestled with the rights and wrongs of including some chart types, unquestionably. The radar chart, for example, has many limitations and flaws but is not entirely without merit if deployed in a very specific way and only for certain contexts. By including profiles of partially flawed charts like these I am using the gallery as much to signpost their shortcomings so that you know to use them sparingly. There will be some purists gathering in angry mobs and foaming at the mouth in reaction to the audacity of my including the pie chart and word cloud. These have limited roles, absolutely, but a role nonetheless. Put down your pitchforks, return to your homes and have a good read of my caveats. Rather than being the poacher of all bad stuff, I think a gamekeeper role is equally important. Although I have excluded several charts on grounds of demonstrating only a slight variation on profiled charts, there are some types included that do exhibit only small derivations from other charts (such as the bar chart and the clustered bar, or the scatter plot and the bubble plot). In these cases I felt there was sufficient difference in their practical application, and they were in common usage, to merit their separate inclusion, despite sharing many similarities with other profiled siblings.
‘Interestingly, visualisations of textual data are not as developed as one
238
would expect. There is a great need for such visualisations given the amount of textual information we generate daily, from social media to news media and so on, not to mention all the materials generated in the past and that are now digitally available. There are opportunities to contribute to the research efforts of humanists as well as social scientists by devising ways to represent not only frequencies of words and topics, but also semantic content. However, this is not at all trivial.’ Isabel Meirelles, Professor, OCAD University (Toronto), discussing one of the many remaining unknowns in visualisation
Categorical comparisons: All chart types can feasibly facilitate comparisons between categories, so why have a separate C family? Well, the distinction is that those charts belonging to the H, R, T and S families offer an additional dimension of analysis as well as providing comparison between categories. Dual families: Some charts do not fit just into a single family. Showing connected relationships (e.g. routes or flows) on a map is ticking the requirements across at least two or family groups (Relational, Spatial). In each case I have tried to best-fit the family classifications around the primary angle of analysis portrayed by each chart – what is the most prominent aspect that characterises each representation technique. Text visualisation: As I noted in the discussion about data types, when it comes to working with textual-based data you are almost always going to need to perform some transformation, maybe through value extraction or by applying a statistical technique. The text itself can otherwise largely function only as an annotated device. Chart types used to visualise text actually visualise the properties of text. For example, the word cloud visualises the quantitative frequency of the use of words: text might be the subject, but categories (words) and their quantities (counts) are the data mappings. Varieties of network diagrams might show the relationship between word usage, such as the sequence of words used in sentences (word trees), but these are still only made possible through some quantitative, categorical or semantic property being drawn from the original text. Dashboard: These methods are popular in corporate settings or any context where you wish to create instrumentation that offers both at- a-glance and detailed views of many different analytical and information monitoring dimensions. Dashboards are not a unique chart type themselves but rather should be considered projects that comprise multiple chart types from across the repertoire of options
239
presented in the gallery. Some of the primary demands of designing dashboards concern editorial thinking (what angles to show and why) and composition choices (how to get it all presented in a unified page layout). Small multiples: This is an invaluable technique for visualising data but not necessarily a chart type per se and, once again, more a concern for about editorial thinking and composition design. Small multiples involve repeated display of the same chart type but with adjustments to the framing of the data in each panel. For example, each panel may show the same angle of analysis but for different categories or different points in time. Small multiples are highly valued because they exploit the capabilities of our visual perception system when it comes to comparing charts in a simultaneous view, overcoming our weakness at remembering and recalling chart views when consumed through animated sequences or across different pages. A note about ‘storytelling’: Storytelling is an increasingly popular term used around data visualisation but I feel it is often misused and misunderstood, which is quite understandable as we all have different perspectives. I also feel it is worth clarifying my take on what I believe storytelling means practically in data visualisation and especially in this discussion about data representation, which is where it perhaps most logically resides in terms of how it is used. Stories are constructs based on the essence of movement, change or narrative. A line chart shows how a series of values have changed over a temporal plane. A flow map can reveal what relationships exist across a spatial plane between two points separated by distance – they may be evident of a journey. However, aside from the temporal and spatial families of charts, I would argue that no other chart family realistically offers this type of construct in and of itself. The only way to create a story from other types of charts is to incorporate a temporal dimension (video/slideshow) or provide a verbal/written narrative that itself involves a dimension of time through the sequence of its delivery. For example, a bar chart alone does not represent a story, but if you show a ‘before’ and ‘after’ pair of bar charts side by side or between slides, you have essentially created ‘change’ through sequence. If you show a bar chart with a stack on top of it to indicate growth between two points in time, well, you have added a time dimension. A network diagram shows relationships, but stood alone this is not a
240
story – its underlying structure and arrangement are in abstract space. Just as you do when showing friends a photograph from your holiday, you might use this chart as a prop to explain how relationships between some of the different entities presented are significant. Making the chart a prop allows you to provide a narrative. In this case it is the setting and delivery that are consistent with the notion of storytelling, not the chart itself. I made a similar observation about the role of exhibitory visualisations used as props within explanatory settings. A further distinction to make is between stories as being presented and stories as being interpreted. The famous six-word story ‘for sale: baby shoes, never worn’ by Ernest Hemingway is not presented as a story, the story is triggered in our mind when we dissect this passage and start to infer meaning, implication and context. The imagined bar chart I mentioned earlier in the book that could show the 43 white presidents and 1 black president is only presenting a story if it is accompanied by an explanatory narrative (in which case the chart was again really just a prop) or if you understand the meaning of the significance of this statistic without this description and are able to form the story in your own mind.
Charts Comparisons
Bar chart
ALSO KNOWN AS Column chart, histogram (wrongly)
REPRESENTATION DESCRIPTION
A bar chart displays quantitative values for different categories. The chart comprises line marks (bars) – not rectangular areas – with the size attribute (length or height) used to represent the quantitative value for each category.
241
EXAMPLE Comparing the number of Oscar nominations for the 10 actors who have received the most nominations without actually winning an award.
Figure 6.8 The 10 Actors with the Most Oscar Nominations but No Wins
HOW TO READ IT & WHAT TO LOOK FOR
Look at the axes so you know with which categorical value each bar is associated and what the range of the quantitative values is (min to max). Think about what high and low values mean: is it ‘good’ to be large or small? Glance across the entire chart to locate the big, small and medium bars and perform global comparisons to establish the high- level ranking of biggest > smallest. Identify any noticeable exceptions and/or outliers. Perform local comparisons between neighbouring bars, to identify larger than and smaller than relationships and estimate the relative proportions. Estimate (or read, if labels are present) the absolute values of specific bars of interest. Where available, compare the quantities against annotated references such as targets, forecast, last year, average, etc.
PRESENTATION TIPS
ANNOTATION: Chart apparatus devices like tick marks and gridlines, in particular, can be helpful to increase the accuracy of the reading of the quantitative values. If you have axis labels you should not need direct labels on each bar – this will lead to label overload, so generally decide between one or the other.
242
COMPOSITION: The quantitative value axis should always start from the origin value of zero: a bar should be representative of the true, full quantitative value, nothing more, nothing less, otherwise the perception of bar sizes will be distorted when comparing relative sizes. There is no significant difference in perception between vertical or horizontal bars though horizontal layouts tend to make it easier to accommodate and read the category labels for each bar. Unlike the histogram, there should be a gap, even if very small, between bars to keep each category’s value distinct. Where possible, try to make the categorical sorting meaningful.
VARIATIONS & ALTERNATIVES
A variation in the use of bar charts is to show changes over time. You would use a bar chart when the focus is on individual quantitative values over time rather than (necessarily) the trend/change between points, for which a line-chart would be best. ‘Spark bars’ are mini bar charts that aim to occupy only a word’s length amount of space. They are often seen in dashboards where space is at a premium and there is a desire to optimise the density of the display. To show further categorical subdivisions, you might consider the ‘clustered bar chart’ or a ‘stacked bar chart’ if there is a part-to-whole angle. ‘Dot plots’ offer a particularly useful alternative to the bar chart for situations where you have to show large quantitative values with a narrow range of differences.
Charts Comparisons
Clustered bar chart
ALSO KNOWN AS Clustered column chart, paired bar chart
REPRESENTATION DESCRIPTION
A clustered bar chart displays quantitative values for different major categories with additional categorical dimensions included for further
243
breakdown. The chart comprises line marks (bars) – not rectangular areas – with the size attribute (length or height) used to represent the quantitative value for each category and colours used to distinguish further categorical dimensions.
EXAMPLE Comparing the number of Oscar nominations with the number of Oscar awards for the 10 actors who have received the most nominations.
Figure 6.9 The 10 Actors who have Received the Most Oscar Nominations
HOW TO READ IT & WHAT TO LOOK FOR
Look at the axes so you know with which categorical value each bar is associated and what the range of the quantitative values is (min to max). Learn about the colour associations to understand what sub-categories the bars within each cluster represent. Glance across the entire chart to locate the big, small and medium bars and perform global comparisons to establish the high-level ranking of biggest > smallest. Identify any noticeable exceptions and/or outliers. Perform local comparisons within clusters to identify the size relationship (which is larger and by how much?) and estimate (or read, if labels are present) the absolute values of specific bars of interest.
PRESENTATION TIPS
ANNOTATION: Chart apparatus devices like tick marks and gridlines,
244
in particular, can be helpful to increase the accuracy of the reading of the quantitative values. If you have axis labels you should not need direct labels on each bar – this will lead to label overload, so generally decide between one or the other.
COMPOSITION: The quantitative value axis should always start from the origin value of zero: a bar should be representative of the true, full quantitative value, nothing more, nothing less, otherwise the perception of bar sizes will be distorted when comparing relative sizes. If your categorical clusters involve a breakdown of more than three bars, it becomes a little too busy, so you might therefore consider giving each cluster its own separate bar chart and using small multiples to show a chart for each major category. Sometimes one bar might be slightly hidden behind the other, implying a before and after relationship, often when space is at a premium – just do not hide too much of the back bar. There is no significant difference in perception between vertical or horizontal bars though horizontal layouts tend to make it easier to accommodate and read the category labels for each bar. The individual bars should be positioned adjacent to each other with a noticeable gap and then between each cluster to help direct the eye towards the clustering patterns first and foremost. Where possible try to make the categorical sorting meaningful.
VARIATIONS & ALTERNATIVES
Clustered bar charts are also sometimes used to show how two associated sub-categories have changed over time (like the Lionel Messi bar chart discussed in Chapter 1). Alternatives would include the ‘dot plot’ or, if you have just two categories forming the clusters and these categories have a binary state (male, female or yes %, no %), the ‘back-to-back bar chart’ would be effective.
Charts Comparisons
Dot plot
245
ALSO KNOWN AS Dot chart
REPRESENTATION DESCRIPTION
A dot plot displays quantitative values for different categories. In contrast to the bar chart, rather than using the size of a bar, point marks (typically circles but any ‘symbol’ is legitimate) are used with the position along a scale indicating the quantitative value for each category. Sometimes an area mark is used to indicate one value through position and another value through size. Additional categorical dimensions can be accommodated in the same chart by including additional marks differentiated by colour or symbol.
EXAMPLE Comparing the number and percentage of PhDs awarded by gender across different academic subjects.
Figure 6.10 How Nations Fare in PhDs by Sex
HOW TO READ IT & WHAT TO LOOK FOR
For single-series dot plots (i.e. just one dot per row), look at the axes so you know with which categorical value each row is associated and what the range of the quantitative values is (min to max). Where you have multiple series dot plots (i.e. more than one dot), establish what the different colours/symbols represent in terms of categorical breakdown. Glance across the entire chart to locate the big, small and medium values and perform global comparisons to establish the high-level ranking of biggest > smallest. Identify any noticeable exceptions and/or
246
outliers. Where you have multiple series look across each series of dot values separately and then perform local comparisons within rows to identify the relative position of each dot, observing the gaps, big and small. Estimate the absolute values of specific dots of interest. Where available, compare the quantities against annotated references such as targets, forecast, last year, average, etc.
PRESENTATION TIPS
ANNOTATION: Chart apparatus devices like tick marks and gridlines, in particular, can be helpful to increase the accuracy of the reading of the quantitative values.
COMPOSITION: Given that the quantitative value axis does not need to commence from a zero origin it is important to label clearly the axis values when the baseline is not commencing from a minimum of zero. There is no significant difference in perception between vertical or horizontal arrangement though horizontal layouts tend to make it easier to accommodate and read the category labels for each row. Where possible try to make the categorical sorting meaningful, maybe organising values in ascending/descending size order.
VARIATIONS & ALTERNATIVES
Alternatives would include the ‘bar chart’, to show the size of quantitative values for different categories. The ‘connected dot plot’ would be used to focus on the difference between two measures. The ‘univariate scatter plot’ would be used to show the range of multiple values across categories, to display the diversity and distribution of values.
Charts Comparisons
Connected Dot Plot
247
ALSO KNOWN AS Barbell chart, dumb-bell chart
REPRESENTATION DESCRIPTION
A connected dot plot displays absolute quantities and quantitative differences between two categorical dimensions for different major categories. The display is formed by two points (normally circles but any ‘symbol’ is legitimate) to mark the quantitative value positions for two comparable categorical dimensions. There is a row of connected dots for each major category. Colour or difference in symbol is generally used to distinguish these points. Joining the two points together is a connecting line which effectively represents the ‘delta’ (difference) between the two values.
EXAMPLE Comparing the typical salaries for women and men across a range of different job categories in the US.
Figure 6.11 Gender Pay Gap US
HOW TO READ IT & WHAT TO LOOK FOR
Look at the axes so you know with which major categorical values each row is associated and what the range of the quantitative values is (min to max). Determine which dots resemble which categorical dimension (could be colour, symbol or a combination) and see if there is any meaning behind the colouring of the connecting bars. Think about what the quantitative values mean to determine whether it is a good thing to be higher or lower. Glance across the entire chart to locate the big,
248
small and medium connecting bars in each direction. Perform global comparisons to establish the high-level ranking of biggest > smallest differences as well as the highest and lowest values. There may be deliberate sorting of the display based on one of the quantitative measures. Identify any noticeable exceptions and/or outliers. Estimate (or read, if labels are present) the absolute values, direction and size of differences for specific categories of interest.
PRESENTATION TIPS
ANNOTATION: Chart apparatus devices like tick marks and gridlines, in particular, can be helpful to increase the accuracy of the reading of the quantitative values. Consider labelling categories adjacent to the plotted points rather than next to the axis line (and possibly far away from the values) to make it easier for the reader to understand the category–row association.
COLOUR TIPS: Colour may be used to indicate and emphasise the directional basis of the connecting line differences.
COMPOSITION: If the two plotted measures are very similar, and the point markers effectively overlap, you will need to decide which should be positioned on top. As the representation of the quantitative values is through position along a scale and not size (it is the difference that is sized, not the absolutes) the quantitative axis does not need to have a zero origin. However, a zero origin can be helpful to establish the scale of the differences. Where possible try to make the sorting meaningful using any one of the three quantitative measures to optimise the layout.
VARIATIONS & ALTERNATIVES
Variations in the use of the ‘connected dot plot’ would show before and after analysis between two points in time, possibly using the ‘arrow chart’ to indicate the direction of change explicitly. Similarly, the ‘carrot chart’ uses line width tapering to indicate direction, the fatter end the more recent values. The ‘univariate scatter plot’ would be used to show the range of multiple values across categories, to display the diversity and distribution of values rather than comparing differences between values.
Charts Comparisons
249
Pictogram
ALSO KNOWN AS Isotype chart, pictorial bar chart, stacked shape chart, tally chart
REPRESENTATION DESCRIPTION
A pictogram displays quantitative values for different major categories with additional categorical dimensions included for further breakdown. In contrast with the bar chart, rather than using the size of a bar, quantities of point marks, in the form of symbols or pictures, are stacked to represent the quantitative value for each category. Each point may be representative of one or many quantitative units (e.g. a single shape may represent 1000 people) but note that, unless you use symbol portions, you will not be able to represent decimals. Pictograms may be used to offer a more emotive (humanising or more light-hearted) display than a bar can offer. Additional categorical dimensions can be accommodated in the same chart by using marks differentiated by variations in colour, symbol or picture. Always ensure the markers used are as intuitively recognisable as possible and consider minimising the variety as this makes it cognitively harder for the viewer to identify associations easily and make sense of the quantities.
EXAMPLE Comparing the number of players with different facial hair types across the four teams in the NHL playoffs in 2015.
Figure 6.12 Who Wins the Stanley Cup of Playoff Beards?
250
HOW TO READ IT & WHAT TO LOOK FOR
Look at the major categorical axis to establish with which category each row is associated. Establish the mark associations to understand what categorical dimensions each colour/shape variation represents. Glance across the entire chart to locate the big, small and medium stacks of shapes and perform global comparisons to establish the high-level ranking of biggest > smallest. Identify any noticeable exceptions and/or outliers. Perform local comparisons between neighbouring categories, to identify larger than and smaller than relationships and estimate the relative proportions. Estimate (or read, if labels are present) the absolute values of specific groups of markers of interest.
PRESENTATION TIPS
ANNOTATION: The choice of symbol/ picture should be as recognisably intuitive as possible and locate any legends as close as possible to the display.
COLOUR TIPS: Maximise the variation in marker by using different combinations in both colour and shape, rather than just variation of one attribute.
COMPOSITION: If the quantities of markers exceed a single row, try to make the number of units per row logically ‘countable’, such as displaying in groups of 5, 10 or 100. To aid readability, make sure there is a sufficiently noticeable gap between rows, otherwise sometimes the eye struggles to form the distinct clusters of shapes for each category displayed. Where possible try to make the categorical sorting meaningful, maybe organising values in ascending/descending size order.
251
VARIATIONS & ALTERNATIVES
Extending the idea of using repeated quantities of representative symbols, some applications take this further by using large quantities of individual symbols to get across the feeling of magnitude and scale. When showing a part-to-whole relationship, the ‘waffle chart’ can use simple symbol devices to differentiate the constituent parts of a whole.
Charts Comparisons
Proportional shape chart
ALSO KNOWN AS Area chart (wrongly)
REPRESENTATION DESCRIPTION
A proportional shape chart displays quantitative values for different categories. The chart is based on the use of different area marks, one for each category, sized in proportion to the quantities they represent. By using the quadratic dimension of area size rather than the linear dimension of bar length or dot position, the shape chart offers scope for displaying a diverse range of quantitative values within the same chart. Typically the layout is quite free-form with no baseline or central gravity binding the display together.
EXAMPLE Comparing the market capitalisation ($) of companies involved in the legal sale of marijuana across different industry sectors.
Figure 6.13 For These 55 Marijuana Companies, Every Day is 4/20
252
HOW TO READ IT & WHAT TO LOOK FOR
Look at the shapes and their associated labels so you know with what major categorical values each is associated. If there are only direct labels, find the largest shape to establish its quantitative value as the maximum and do likewise for the smallest – this will help calibrate the size judgements. Otherwise, if it exists, acquaint yourself with the size key. Glance across the entire chart to locate the big, small and medium shapes and perform global comparisons to establish the high-level ranking of biggest > smallest. Identify any noticeable exceptions and/or outliers. Perform local comparisons between neighbouring shapes to identify larger than and smaller than relationships and estimate the relative proportions. Estimate (or read, if labels are present) the absolute values of specific shapes of interest.
PRESENTATION TIPS
ANNOTATION: Sometimes a quantitative size key will be included rather than direct labelling (usually when there are many shapes and limited empty space) though direct labels will help overcome some of the limitations of judging area size. You will have to decide how to handle label positioning for those shapes with exceptionally small sizes.
COLOUR TIPS: Colours are not fundamentally necessary to encode category (the position/separation of different shapes achieves that already) but they can be useful as redundant encodings to make the category even more immediately distinguishable.
COMPOSITION: Estimating and comparing the size of areas with
253
accuracy is not as easy as it is for judging bar length or dot position, so only use this chart type if you have a diverse range of quantitative values. The geometric accuracy of the size calculations is paramount. Mistakes are often made, in particular, with circle size calculations: it is the area you are modifying, not the diameter/radius. Arrangement approaches vary: sometimes you see the shapes anchored to a common baseline (bottom or central alignment) while on other occasions they might just ‘float’. If you use an organic shape, like a human figure, to represent different quantities you need to adjust the entire shape area, not just the height. Often the approach for this type of display is to treat the figure as a rudimentary rectangular shape. Sometimes the volume of a shape is used rather than area to represent quantitative values (especially if there are almost exponentially different values to show) but this increases the perceptual difficulty in estimating and comparing values. Where possible try to make the categorical sorting meaningful, maybe organising values in ascending/descending size order.
VARIATIONS & ALTERNATIVES
The ‘bubble chart’ uses clusters of sized bubbles to compare categorical values and, sometimes, to represent part-to-whole analysis. The ‘nested shape chart’ might include secondary, smaller area sizes nested within each shape to display local part-to-whole relationships.
Charts Comparisons
Bubble chart
ALSO KNOWN AS Circle packing diagram
EXAMPLE Comparing the Public sector capital expenditure (£ million) on services by function of the UK Government during 2014/15.
REPRESENTATION DESCRIPTION
254
A bubble chart displays quantitative values for different major categories with additional categorical dimensions included for further breakdown. It is based on the use of circles, one for each category, sized in proportion to the quantities they represent. Sometimes several separate clusters may be used to display further categorical dimensions, otherwise the colouring of each circle can achieve this. It is similar in concept to the proportional shape chart but differs through the typical layout being based on clustering, which therefore also enables it as a device for showing part-to-whole relationships as well.
Figure 6.14 UK Public Sector Capital Expenditure, 2014/15
HOW TO READ IT & WHAT TO LOOK FOR
Look at the shapes and their associated labels so you know with what major categorical values each is associated, noting any size and colour legends to assist in forming associations. If there are multiple clusters, learn about the significance of the grouping/separation in each case. If there are direct labels, find the largest shape to establish its quantitative value as the maximum and do likewise for the smallest – this will help calibrate other size judgements. Glance across the entire chart to locate the big, small and medium shapes and perform global comparisons to establish the high-level ranking of biggest > smallest. Identify any noticeable exceptions and/or outliers. Perform local comparisons between neighbouring shapes to identify larger than and smaller than relationships and estimate the relative proportions. Estimate (or read, if labels are present) the absolute values of specific shapes of interest. If there are multiple clusters, note the general relative size and number of members in each case.
255
PRESENTATION TIPS
INTERACTIVITY: Bubble charts may often be accompanied by interactive features that let users select or mouseover individual circles to reveal annotated values for the quantity and category.
ANNOTATION: If interactivity is not achievable, a quantitative size key should be included or direct labelling; the latter may make the display busy (and be hard to fit into smaller circles) but will help overcome some of the limitations of judging area size.
COLOUR TIPS: Colours are sometimes used as redundant encodings to make the quantitative sizes even more immediately distinguishable.
COMPOSITION: Estimating and comparing the size of areas with accuracy is not as easy as it is for judging bar length or dot position, so only use this chart type if you have a diverse range of quantitative values. The use of this chart will primarily be about facilitating a gist, a general sense of the largest and smallest values. The geometric accuracy of the circle size calculations is paramount. Mistakes are often made with circle size calculations: it is the area you are modifying, not the diameter/radius. If you wish to make your bubbles appear as 3D spheres you are essentially no longer representing quantitative values through the size of a geometric area mark; rather the mark will be a ‘form’ and so the size calculation will be based on volume, not area. There is no categorical or quantitative sorting applied to the layout of the bubble chart, instead the tools that offer these charts will generally use a layout algorithm that applies a best-fit clustering to arrange the circles radially about a central ‘gravity’ force.
VARIATIONS & ALTERNATIVES
When the collection of quantities represents a whole, this evolves into a chart known as a ‘circle packing diagram’ and usually involves many parts that pack neatly into a circular layout representing the whole. Another variation of the packing diagram is when the adjacency between circle ‘nodes’ indicates a connected relation, offering a variation of the node–link diagram for showing networks of relationships. The bubble plot also uses differently sized circles but the position in each case is overlaid onto a scatter plot structure, based on two dimensions of further quantitative variables. Removing the size attribute (and effectively replacing area with point mark) you could simply use the quantity of points clustered together for different categories to create a ’tally chart’.
256
Charts Comparisons
Radar chart
ALSO KNOWN AS Filled radar chart, star chart, spider diagram, web chart
EXAMPLE Comparing the global competitive scores (out of 7) across 12 ‘pillars’ of performance for the United Kingdom.
REPRESENTATION DESCRIPTION
A radar chart shows values for three or more different quantitative measures in the same display for, typically, a single category. It uses a radial (circular) layout comprising several axes emerging from the centre-like spokes on a wheel, one for each measure. The quantitative values for each measure are plotted through position along each scale and then joined by connecting lines to form a unique geometric shape. Sometimes this shape is then filled with colour. A radar chart should only be considered in situations where the cyclical ordering (and neighbourly pairings) has some significance (such as data that might be plotted around the face of a clock or compass) and when the quantitative scales are the same (or similar) for each axis. Do not plot values for multiple categories on the same radar chart, but use small multiples formed of several radar charts instead.
Figure 6.15 Global Competitiveness Report 2014—2015
257
HOW TO READ IT & WHAT TO LOOK FOR
Look around the chart and acquaint yourself with the quantitative measure represented by each axis and note the sequencing of the measures around the display. Is there any significance in this arrangement that can assist in interpreting the overall shape? Note the range of values along each independent axis so you understand what positions along the scales mean in a value sense for each measure. Scan the shape to locate the outliers both towards the outside (larger values) and inside (smaller values) of the scales. It is more important to pay attention to the position of values along an axis than the nature of the connecting lines between axes, unless the axis scales are consistent or at least if the relative position along the scale has the same implied meaning. If the variable sequencing has cyclical relevance, the spiking, bulging or contracting shape formed will give you some sense of the balance of values. Perform local comparisons between neighbouring axes to identify larger than and smaller than relationships. Estimate (or read, if labels are present) the absolute values of specific shapes of interest.
PRESENTATION TIPS
258
ANNOTATION: The inclusion of visible annotated features like axis lines, tick marks, gridlines and value labels can naturally aid the readability of the radar chart. Gridlines are only relevant if there are common scales across each quantitative variable. If so, the gridlines must be presented as straight lines, not concentric arcs, because the connecting lines joining up the values are themselves straight lines.
COLOUR TIPS: Often the radar shapes are filled with a colour, sometimes with a degree of transparency to allow the background apparatus to be partially visible.
COMPOSITION: The cyclical ordering of the quantitative variables has to be of optimum significance as the connectors and shape change for every different ordering permutation. This will have a major impact on the readability and meaning of the resulting chart shape. As the axes will be angled all around the radial display, you will need to make sure all the associated labels are readable (i.e. not upside down or at difficult angles).
VARIATIONS & ALTERNATIVES
A ‘polar chart’ is an alternative to the radar chart that removes some of the main shortcomings caused by connecting lines in the radar chart. If you have consistent value scales across the different quantitative measures, a ‘bar chart’ or ‘dot plot’ would be a better alternative. While not strictly a variation, ‘parallel coordinates’ display a similar technique for plotting several independent quantitative measures in the same chart. The main difference is that parallel coordinates use a linear layout and can accommodate many categories in one display.
Charts Comparisons
Polar chart
ALSO KNOWN AS Coxcomb plot, polar area plot
259
REPRESENTATION DESCRIPTION
A polar chart shows values for three or more different quantitative measures in the same display. It uses a radial (circular) layout comprising several equal-angled circular sectors like slices of a pizza, one for each measure. In contrast to the radar chart (which uses position along a scale), the polar chart uses variation in the size of the sector areas to represent the quantitative values. It is, in essence, a radially plotted bar chart. Colour is an optional attribute, sometimes used visually to indicate further categorical dimensions. A polar chart should only be considered in situations where the cyclical ordering (and neighbourly pairings) has some significance (such as data that might be plotted around the face of a clock or compass) and when the quantitative scales are the same (or similar) for each axis.
EXAMPLE Comparing the quantitative match statistics across 14 different performance measures for a rugby union player.
Figure 6.16 Excerpt from a Rugby Union Player Dashboard
HOW TO READ IT & WHAT TO LOOK FOR
260
Look around the chart and acquaint yourself with the quantitative measures each sector represents and note the sequencing of the measures around the display. Is there any significance in this arrangement that can assist in interpreting the overall shape? Note the range of values included on the quantitative scale and acquaint yourself with any colour associations. Glance across the entire chart to locate the big, small and medium sectors and perform global comparisons to establish the high-level ranking of biggest > smallest. Identify any noticeable exceptions and/or outliers. Perform local comparisons between neighbouring variables to identify the order of magnitude and estimate the relative sizes. Estimate (or read, if labels are present) the absolute values of specific sectors of interest. Where available, compare the quantities against annotated references such as targets, forecast, last year, average, etc. If there is significance behind the sequencing of the variables, look out for any patterns that emerge through spiking, bulging or contracting shapes.
PRESENTATION TIPS
ANNOTATION: The inclusion of visible annotated features like tick marks and value labels can naturally aid the readability of the polar chart. Gridlines are only relevant if there are common scales across each quantitative variable. If so, the gridlines must be presented as arcs reflecting the outer shape of each sector. Connecting lines joining up the values are themselves straight lines. Each sector typically uses the same quantitative scale for each quantitative measure but, on the occasions when this is not the case, each axis will require its own, clear value scale.
COLOUR TIPS: Often polar chart sectors are filled with a meaningful colour, sometimes with a degree of transparency to allow the background apparatus to be partially visible.
COMPOSITION: The cyclical ordering of the quantitative variables has to be of some significance to legitimise the value of the polar chart over the bar chart. As the sectors will be angled all around the radial display, you will need to make sure all the associated labels are readable (i.e. not upside down or at difficult angles). The quantitative values represented by the size of the sectors need to be carefully calculated. It is the area of the sector, not the radius length, that will be modified to portray the values accurately. If you make maximum quantitative value equivalent to the largest sector area, all other sector sizes can be calculated accordingly. Knowing how many different quantitative variables you are showing means you can easily calculate
261
the angle of any given sector. The quantitative measure axes should always start from the origin value of zero: a sector should be representative of the true, full quantitative value, nothing more, nothing less, otherwise the perception of size will be distorted when comparing relative sizes.
VARIATIONS & ALTERNATIVES
Unless the radial layout provides meaning through the notion of a ‘whole’ or through the cyclical arrangement of measures, you might be best using a ‘bar chart’. Variations in approach tend to see modifications in the sector shape with measure values represented by individual bars lengths or, in the example of the Better Life Index project, through variations in ‘petal’ sizes.
Charts Distributions
Range Chart
ALSO KNOWN AS Span chart, floating bar chart, barometer chart
REPRESENTATION DESCRIPTION
A range chart displays the minimum to maximum distribution of a series of quantitative values for different categories. The display is formed by a bar, one for each category, with the lower and upper position of the bars shaped by the minimum and maximum quantitative values in each case. The resulting bar lengths thus represent the range of values between the two limits.
EXAMPLE Comparing the highest and lowest temperatures (°F) recorded across the top 10 most populated cities during 2015.
Figure 6.17 Range of Temperatures Recorded in Top 10 Most Populated Cities (2015)
262
HOW TO READ IT & WHAT TO LOOK FOR
Look at the axes so you know with what major categorical values each range bar is associated and what the range of the quantitative values is (min to max). Glance across the entire chart to locate the big, small and medium bars and perform global comparisons to establish the high- level ranking of biggest > smallest differences as well as the highest and lowest values. Identify any noticeable exceptions and/or outliers. Perform local comparisons between neighbouring bars, to identify larger than and smaller than relationships and estimate the relative proportions. There may be deliberate sorting of the display based on one of the quantitative measures. Estimate (or read, if labels are present) the absolute values of specific bars of interest. Where available, compare the quantities against annotated references such as targets, forecast, last year, average, etc.
PRESENTATION TIPS
ANNOTATION: Chart apparatus devices like tick marks and gridlines, in particular, can be helpful to increase the accuracy of the reading of the quantitative values. If you have axis labels you may not need direct labels on each bar – this will be lead to label overload, so generally decide between one or the other.
COMPOSITION: The quantitative value axis does not need to commence from zero, unless it means something significant to the
263
interpretation, as the range of values themselves does not necessarily start from zero and the focus is more on the range and difference between the outer values. There is no significant difference in perception between vertical or horizontal layouts, though the latter tend to make it easier to accommodate and read the category labels. Where possible, try to make the categorical sorting meaningful, maybe organising values in ascending/descending size order.
VARIATIONS & ALTERNATIVES
‘Connected dot plots’ will also emphasise the difference between two selected measure values (as opposed to min/max) or where the underlying data is a change over time between two observations. ‘Band charts’ will often be used to show how the range of data values has changed over time, displaying the minimum and maximum bands at each time unit. These are often used in displays like weather forecasts.
Charts Distributions
Box-and-whisker plot
ALSO KNOWN AS Box plot
REPRESENTATION DESCRIPTION
A box-and-whisker plot displays the distribution and shape of a series of quantitative values for different categories. The display is formed by a combination of lines and point markers to indicate (through position and length), typically, five different statistical measures. Three of the statistical values are common to all plots: the first quartile (25th percentile), the second quartile (or median) and the third quartile (75th percentile) values. These are displayed with a box (effectively a wide bar) positioned and sized according to the first and third quartile values with a marker indicating the median. The remaining two statistical values vary in definition: usually either the minimum and maximum
264
values or the 10th and 90th percentiles. These statistical values are represented by extending a line beyond the bottom and top of the main box to join with a point marker indicating the appropriate position. These are the whiskers. A plot will be produced for each major category.
EXAMPLE Comparing the distribution of annual earnings 10 years after starting school for graduates across the eight Ivy League schools.
Figure 6.18 Ranking the Ivies
HOW TO READ IT & WHAT TO LOOK FOR
Begin by looking at the axes so you know with which category each plot is associated and what the range of quantitative values is (min to max). Establish the specific statistics being displayed, by consulting any legends or descriptions, especially in order to identify what the ‘whiskers’ are representing. Glance across the entire chart to locate the main patterns of spread, identifying any common or noticeably different
265
patterns across categories. Look across the shapes formed for each category to learn about the dispersal of values: starting with the median, then observing the extent and balance of the ‘box’ (the interquartile range between the 25th and 75th percentiles) and then check the ‘whisker’ extremes. Is the shape balanced or skewed around the median? Is the interquartile range wide or narrow? Are the whisker extremes far away from the edges of the box? Then return to comparing shapes across all categories to identify more precisely any interesting differences or commonalities for each of the five statistical measures.
PRESENTATION TIPS
ANNOTATION: If you have axis labels you may not need direct labels on each bar – this will lead to label overload, so generally decide between one or the other.
COMPOSITION: The quantitative value axis does not need to commence from zero, unless it means something significant to the interpretation, as the range of values themselves do not necessarily start from zero and the focus is on the statistical properties between the outer values. There is no significant difference in perception between vertical or horizontal box-and-whisker plots, though horizontal layouts tend to make it easier to accommodate and read the category labels. Try to keep a noticeable gap between plots to enable greater clarity in reading. When you have several or many plots in the same chart, where possible try to make the categorical sorting meaningful, maybe organising values in ascending/descending order based on the median value.
VARIATIONS & ALTERNATIVES
Variations involve reducing the number of statistical measures included in the display by removing the whiskers to just show the 25th and 75th percentiles through the lower and upper parts of the box. The ‘candlestick chart’ (or OHLC chart) involves a similar approach and is often used in finance to show the distribution and milestone values of stock performances during a certain time frame (usually daily), plotting the opening, highest, lowest and closing prices, using colour to indicate an up or down trend.
Charts Distributions
266
Univariate scatter plot
ALSO KNOWN AS 1D scatter plot, jitter plot
REPRESENTATION DESCRIPTION
A univariate scatter plot displays the distribution of a series of quantitative values for different categories. In contrast to the box-and- whisker plot, which shows selected statistical values, a univariate scatter plot shows all values across a series. For each category, a range of points (typically circles but any ‘symbol’ is legitimate) are used to mark the position along the scale of the quantitative values. From this you can see the range, the outliers and the clusters and form an understanding about the general shape of the data.
EXAMPLE Comparing the distribution of average critics score (%) from the Rotten Tomatoes website for each movie released across a range of different franchises and movie theme collections.
Figure 6.19 Comparing Critics Scores for Major Movie Franchises
267
HOW TO READ IT & WHAT TO LOOK FOR
Look at the axes so you know what each scatter row/column relates to in terms of which category it is associated with and what the range of the quantitative values is (min to max). If colour has been used to emphasise or separate different marks, establish what the associations are. Also, learn about how the design depicts multiple marks on the same value – these may appear darker or indeed larger. Glance across the entire chart to observe the main patterns of clustering and identify any noticeable exceptions and/or outliers across all categories. Then look more closely at the patterns within each scatter to learn about each category’s specific dispersal of values. Look for empty regions where no quantitative values exist. Estimate the absolute values of specific dots of interest. Where available, compare the quantities against annotated references such as the average or median.
PRESENTATION TIPS
ANNOTATION: Chart apparatus devices like gridlines can be helpful to increase the accuracy of the reading of the quantitative values. Direct labelling is normally restricted to including values for specifically noteworthy points only.
COLOUR: Colour may be used to establish focus of certain points
268
and/or distinction between different sub-category groups to assist with interpretation. When several points have the exact same value you might need to use unfilled or semi-transparent filled circles to facilitate a sense of value density.
COMPOSITION: The representation of the quantitative values is based on position and not size, therefore the quantitative axis does not need to have a zero origin. There is no significant difference in perception between vertical or horizontal arrangement, though horizontal layouts tend to make it easier to accommodate and read the category labels. Where possible try to make the categorical sorting meaningful, maybe organising values in ascending/descending size order.
VARIATIONS & ALTERNATIVES
To overcome occlusion caused by plotting several marks at the same value, a variation of the univariate scatter plot may see the points replaced by geometric areas (like circles), where the position attribute is used to represent a quantitative value along a scale and the size attribute is used to indicate the frequency of observations of similar value. Adding a second quantitative variable axis would lead to the use of a ’scatter plot’.
Charts Distributions
Histogram
ALSO KNOWN AS Bar chart (wrongly)
REPRESENTATION DESCRIPTION
A histogram displays the frequency and distribution for a range of quantitative groups. Whereas bar charts compare quantities for different categories, a histogram technically compares the number of
269
observations across a range of value ‘bins’ using the size of lines/bars (if the bins relate to values with equal intervals) or the area of rectangles (if the bins have unequal value ranges) to represent the quantitative counts. With the bins arranged in meaningful order (that effectively form ordinal groupings) the resulting shape formed reveals the overall pattern of the distribution of observations.
EXAMPLE Comparing the distribution of movies released over time starring Michael Caine across five-year periods based on the date of release in the US.
Figure 6.20 A Career in Numbers: Movies Starring Michael Caine
HOW TO READ IT & WHAT TO LOOK FOR
Begin by looking at the axes so you know what the chart depicts in terms of the categorical bins and the range of the quantitative values (zero to max). Glance across the entire chart to establish the main pattern. Is it symmetrically shaped, like a bell or pyramid (around a median or average value)? Is it skewed to the left or right? Does it dip in the middle and peak at the edges (known as bimodal)? Does it have several peaks and troughs? Maybe it is entirely random in its pattern?
270
All these characteristics of ‘shape’ will inform you about the underlying distribution of the data.
PRESENTATION TIPS
ANNOTATION: Chart apparatus devices like tick marks and gridlines in particular can be helpful to increase the accuracy of the reading of the quantitative values. Axis labels more than direct value labels tend to be used so as not to crowd the shape of the histogram.
COMPOSITION: Unlike the bar chart there should be no (or at most a very thin) gap between bars to help the collective shape of the frequencies emerge. The sorting of the quantitative bins must be in ascending order so that the reading of the overall shape preserves its meaning. The number of value bins and the range of values covered by each have a prominent influence over the appearance of the histogram and the usefulness of what it might reveal: too few bins may disguise interesting nuances, patterns and outliers; too many bins and the most interesting shapes may be abstracted by noise above signal. There is no singular best approach, the right choice simply arrives through experimentation and iteration.
VARIATIONS & ALTERNATIVES
For analysis that looks at the distribution of values across two dimensions, such as the size of populations for age across genders, a ‘back-to-back histogram’ (with male on one side, female on the other), also commonly known as a ‘violin plot’ or ‘population pyramid’, is a useful approach to see and compare the respective shapes. A ‘box-and- whisker plot’ reduces the distribution of values to five key statistical measures to describe key dimensions of the spread of values.
Charts Distributions
Word cloud
271
ALSO KNOWN AS Tag cloud
REPRESENTATION DESCRIPTION
A word cloud shows the frequency of individual word items used in textual data (such as tweets, comments) or documents (passages, articles). The display is based around an enclosed cluster of words with the font (not the word length) sized according to the frequency of usage. In modifying the size of font this is effectively increasing the area size of the whole word. All words have a different shape and size so this can make it quite difficult to avoid the prominence of long words, irrespective of their font size. Word clouds are therefore only useful when you are trying to get a quick and rough sense of some of the dominant keywords used in the text. They can be an option for working with qualitative data during the data exploration stage, more so as a means for reporting analysis to others.
EXAMPLE Comparing the frequency of words used in Chapter 1 of this book.
Figure 6.21 Word Cloud of the Text from Chapter 1
HOW TO READ IT & WHAT TO LOOK FOR
The challenge with reading word clouds is to avoid being drawn to the length and/or area of a word – they are simply attributes of the word, not a meaningful representation of frequency. It is the size of the font that you need to focus on. Scan the display to spot the larger text showing the more frequently used words. Consider any words of specific interest to see if you can find them; if they are not significantly visible, that in itself could be revealing. While most word cloud
272
generators will dismiss many irrelevant words, you might still need to filter out perceptually the significance of certain dominantly sized text.
PRESENTATION TIPS
INTERACTIVITY: Interactive features that let users interrogate, filter and scrutinise the words in more depth, perhaps presenting examples of their usage in a passage, can be quite useful to enhance the value of a word cloud.
ANNOTATION: While the absolutes are generally of less interest than relative comparisons, to help viewers get as much out of the display as possible a simple legend explaining how the font size equates to frequency number can be useful.
COLOUR: Colours may be used as redundant encoding to accentuate further the larger frequencies or categorically to create useful visual separation.
COMPOSITION: The arrangement of the words within a word cloud is typically based on a layout process. Although not random, this will generally prioritise the placement of words to occupy optimum collective space that preserves an overall shape (with essentially a central gravity) over and above any arrangement that might better enable direct comparison.
VARIATIONS & ALTERNATIVES
The alternative approach would be to use any other method in this categorical family of charts that would more usefully display the counts of text, such as a bar chart.
Charts Part-to-whole
Pie chart
273
ALSO KNOWN AS Pizza chart
REPRESENTATION DESCRIPTION
A pie chart shows how the quantities of different constituent categories make up a whole. It uses a circular display divided into sectors for each category, with the angle representing each of the percentage proportions. The resulting size of the sector (in area terms) is a spatial by-product of the angle applied to each part and so offers an additional means for judging the respective values. The role of a pie chart is primarily about being able to compare a part to a whole than being able to compare one part to another part. They therefore work best when there are only two or three parts included. There are a few important rules for pie charts. Firstly, the total percentage values of all sector values must be 100%; if the aggregate is greater than or less than 100% the chart will be corrupted. Secondly, the whole has to be meaningful – often people just add up independent percentages but that is not what a pie chart is about. Finally, the category values must represent exclusive quantities; nothing should be counted twice or overlap across different categories. Despite all these warnings, do not be afraid of the pie chart – just use it with discretion.
EXAMPLE Comparing the proportion of eligible voters in the 2015 UK election who voted for the Conservative Party, for other parties and who did not vote.
Figure 6.22 Summary of Eligible Votes in the UK General Election 2015
274
HOW TO READ IT & WHAT TO LOOK FOR
Begin by establishing which sectors relate to what categories. This may involve referring to a colour key legend or through labels directly adjacent to the pie. Quickly scan the pie to identify the big, medium and small sectors. Notice if there is any significance behind the ordering of the parts. Unless there are value labels, you next will attempt to judge the individual sector angles. This usually involves mentally breaking the pie into 50% halves (180°) or 25% quarters (90°) and using those guides to perceptually measure the category values. Comparing parts against other parts with any degree of accuracy will only be possible once you have formed estimates of the individual sector sizes. If you are faced with the task of judging the size of many parts it is quite understandable if you decide to give up quite soon.
PRESENTATION TIPS
ANNOTATION: The use of local labelling for category values can be useful but too many labels can become cluttered, especially when attempting to label very small angled sectors.
COLOUR: Colour is generally vital to create categorical separation
275
and association of the different sectors so aim to use the difference in colour hue and not colour saturation to maximise the visible difference.
COMPOSITION: Positioning the first slice at the vertical 12 o’clock position gives a useful baseline to help judge the first sector angle value. The ordering of sectors using descending values or ordinal characteristics helps with the overall readability and allocation of effort. Do not consider using gratuitous decoration (like 3D, gradient colours, or exploding slices).
VARIATIONS & ALTERNATIVES
Sometimes a pie chart has a hole in the centre and is known as a ‘doughnut chart’, continuing the food-related theme. The function is exactly the same as a pie but the removal of the centre, often to accommodate a labelling property, removes the possibility of the reader judging the angles at the origin. One therefore has to derive the angles from the resulting arc lengths. If you want to display multiple parts (more than three) the bar chart will be a better option and, for many parts, the ‘treemap’ is best. Depending on the allocated space, a ‘stacked bar chart’ may provide an alternative to the pie. Unlike most chart types, the pie chart does not work well in the form of small multiples (unless there is only a single part being displayed). A ‘nested shape chart’, typically based on embedded square or circle areas, enables comparison across a series of one-part-to-whole relationships based on absolute numbers, rather than percentages, where the wholes may vary in size.
Charts Part-to-whole
Waffle chart
ALSO KNOWN AS Square pie, unit chart, 100% stacked shape chart
REPRESENTATION DESCRIPTION
276
A waffle chart shows how the quantities of different constituent categories make up a whole. It uses a square display usually representing 100 point ‘cells’ through a 10 × 10 grid layout. Each constituent category proportion is displayed through colour-coding a proportional number of cells. Difference in symbol can also be used. The role of the waffle chart is to simplify the counting of proportions in contrast to the angle judgements of the pie chart, though the display is limited to rounded integer values. This is easier when the grid layout facilitates quick recognition of units of 10. As with the pie chart, the waffle chart works best when you are showing how a single part compares to the whole and perhaps offers greater visual impact when there are especially small percentages of a whole. Rather than just colouring in the grid cells, sometimes different symbols will be used to associate with different categories. For example, you might see figures or gender icons used to show the makeup of a given sample population.
EXAMPLE Comparing the proportion of total browser usage for Internet Explorer and Chrome across key milestone moments.
Figure 6.23 The Changing Fortunes of Internet Explorer and Google Chrome
HOW TO READ IT & WHAT TO LOOK FOR
Begin by establishing how the different shapes or colours are associated with different categories. Assess the grid layout to understand the dimension of the chart and the quantity of cell ‘units’ forming the display (e.g. is it a 10 x 10 grid?). Quickly scan the chart to identify the big, medium and small sectors. Notice if there is any significance
277
behind the ordering of the parts. Unless there are value labels, you will need to count/estimate the number of units representing each category value. Comparing parts against other parts will only be possible once you have established the individual part sizes. If several related waffle charts are shown, possibly for different categories or points in time, identify the related colours/shapes in each chart and establish the patterns of size between and across the various charts, looking for trends, declines and general differences.
PRESENTATION TIPS
ANNOTATION: Direct labelling can become very cluttered and hard to incorporate elegantly without the need for long arrows.
COLOUR: Borders around each square cell are useful to help establish the individual units, but do not make the borders too thick to the point where they dominate attention.
COMPOSITION: Always start each row of values from the same side, for consistency and to make it easier for people to estimate the values. When you have several parts in the same waffle chart, where possible try to make the categorical sorting meaningful, maybe organising values in ascending/descending size order or based on a logical categorical order.
VARIATIONS & ALTERNATIVES
Sometimes the waffle chart approach is used to show stacks of absolute unit values and indeed there are overlaps in concept between this variation in the waffle chart and potential applications of the pictogram. Aside from the pie chart, a ‘nested shape chart’ will provide an alternative way of showing a part-to-whole relationship while also occupying a squarified layout.
Charts Part-to-whole
Stacked bar chart
278
ALSO KNOWN AS
REPRESENTATION DESCRIPTION
A stacked bar chart displays a part-to-whole breakdown of quantitative values for different major categories. The percentage proportion of each categorical dimension or ‘part’ is represented by separate bars, distinguished by colour, that are sized according to their proportion and then stacked to create the whole. Sometimes the whole is standardised to represent 100%, at other times the whole will be representative of absolute values. Stacked bar charts work best when the parts are based on ordinal dimensions, which enables ordering of the parts within the stack to help establish the overall shape of the data. If the parts are representative of nominal data, it is best to keep the number of constituent categories quite low, as estimating the size of individual stacked parts when there are many becomes quite hard.
EXAMPLE Comparing the percentage of adults (16–65 year olds) achieving different proficiency levels in literacy across different countries.
Figure 6.24 Literarcy Proficiency: Adult Levels by Country
279
HOW TO READ IT & WHAT TO LOOK FOR
Look at the axes so you know with what major categorical values each bar is associated and what the quantitative values are, determining if it is a 100% stacked bar or an absolute stacked bar (in which case identify the min and the max). Establish the colour association to understand what categories the bars within each stack represent. Glance across the entire chart. If the categorical data is ordinal, and the sorting/colour of the stacks is intuitive, you should be able to derive meaning from the overall balance of colour patterns, especially where any annotated gridlines help to guide your value estimation. If the categorical data is nominal, seek to locate the dominant colours and the least noticeable ones. Comparing across different stacked bars is made harder by the lack of a common baseline for anything other than the bottom stack on the zero baseline (and for 100% stacked bars, those final ones at the top) and so a general sense of magnitude will be your focus. Study closer the constituent parts within each stack to establish the high-level ranking of biggest > smallest. Estimate (or read, if labels are present)
280
the absolute values of specific stacked parts of interest.
PRESENTATION TIPS
ANNOTATION: Direct value labelling can become very cluttered when there are many parts or stacks and you are comparing several different major categories. You might be better with a table if that is your aim. Definitely include value axis labels with logical intervals and it is very helpful to annotate, through gridlines, key units such as the 25%, 50% and 75% positions when based on a 100% stacked bar chart.
COLOUR: If you are representing categorical ordinal data, colour can be astutely deployed to give a sense of the general balance of values within the whole, but this will only work if their sorting arrangement within the stack is logically applied. For categorical nominal data, ensure the stacked parts have sufficiently different colours so that their distinct bar lengths can be efficiently observed.
COMPOSITION: Across the main categories, once again consider the optimum sorting option, maybe organising values in ascending/descending size order or based on a logical categorical order. Judging the size of the stacks with accuracy is harder for those that are not on the zero baseline, so maybe consider which ones are of most importance to be more easily read and place those on the baseline.
VARIATIONS & ALTERNATIVES
The main alternative would be to use ‘multi-panel bar charts’, where separate bar charts each include just one ’stack’/part and they are then repeated for each subsequent constituent category. In the world of finance the ‘waterfall chart’ is a common approach based on a single stacked bar broken up into individual elements, almost like a step-by- step narrative of how the components of income look on one side and then how the components of expenditure look on the other, with the remaining space representing the surplus or deficit. Like their unstacked siblings, stacked bar charts can also be used to show how categorical composition has changed over time.
Charts Part-to-whole
281
Back-to-back bar chart
ALSO KNOWN AS Paired bar chart
REPRESENTATION DESCRIPTION
A back-to-back bar chart displays a part-to-whole breakdown of quantitative values for different major categories. As with any bar chart, the length of a bar represents a quantitative proportion or absolute value for each part and across all major categories. In contrast to the stacked bar chart, where the constituent bars are simply stacked to form a whole, in a back-to-back bar chart the constituent parts are based on diverging categorical dimensions with a ‘directional’ essence such as yes/no, male/female, agree/disagree. The values for each dimension are therefore presented on opposite sides of a shared zero baseline to help reveal the shape and contrast differences across all major categories.
EXAMPLE Comparing the responses to a survey question asking for opinions about ‘the government collection of telephone and Internet data as part of anti-terrorism efforts’ across different demographic categories.
Figure 6.25 Political Polarization in the American Public
282
HOW TO READ IT & WHAT TO LOOK FOR
Look at the axes so you know with which major categorical values each bar is associated and what the range of the quantitative values is (min to max). Establish what categorical dimensions are represented by the respective sides of the display and any colour associations. Glance across the entire chart to locate the big, small and medium bars and perform global comparisons to establish the high-level ranking of biggest > smallest. Repeat this for each side of the display, noticing any patterns of dominance of larger values on either side. Identify any noticeable exceptions and/or outliers. Perform local comparisons for each category value to estimate the relative sizes (or read, if labels are present) of each bar.
PRESENTATION TIPS
ANNOTATION: Chart apparatus devices like tick marks and gridlines in particular can be helpful to increase the accuracy of the reading of the quantitative values.
COLOUR: The bars either side of the axis do not need to be coloured but often are to create further visual association.
COMPOSITION: The quantitative value axis should always start from the origin value of zero: a bar should be representative of the true, full quantitative value, nothing more, nothing less, otherwise the perception
283
of bar sizes will be distorted when comparing relative sizes. There is no significant difference in perception between vertical or horizontal bars, though horizontal layouts tend to make it easier to accommodate and read the category labels. Where possible try to make the categorical sorting meaningful, maybe organising values in ascending/descending size order or based on a logical categorical order.
VARIATIONS & ALTERNATIVES
Back-to-back bar charts facilitate a general sense of the shape of diverging categorical dimensions. However, if you want to facilitate direct comparison, a ‘clustered bar chart’ showing adjacent bars helps to compare respective heights more precisely. For analysis that looks at the distribution values across two dimensions, such as the size of populations for age across genders, a ‘back-to-back histogram’ (with male on one side, female on the other), also commonly known as a ‘violin plot’ or ‘population pyramid’, is a useful approach to see and compare the respective shapes. Some back-to-back applications do not show a part-to-whole relationship but simply compare quantities for two categorical values. Further variations may appear as ‘back-to-back area charts’ showing mutual change over time for two contrasting states.
Charts Part-to-whole
Treemap
ALSO KNOWN AS Heat map (wrongly)
REPRESENTATION DESCRIPTION
A treemap is an enclosure diagram providing a hierarchical display to show how the quantities of different constituent parts make up a whole. It uses a contained rectangular layout (often termed ‘squarified’) representing the 100% total divided into proportionally sized
284
rectangular tiles for each categorical part. Colour can be used to represent an additional quantitative measure, such as an indication of amount of change over a time period. The absolute positioning and dimension of each rectangle is organised by an underlying tiling algorithm to optimise the overall space usage and to cluster related categories into larger rectangle-grouped containers. Treemaps are most commonly used, and of most value, when there are many parts to the whole but they are only valid if the constituent units are legitimately part of the same ‘whole’.
EXAMPLE Comparing the relative value of and the daily performance of stocks across the S&P 500 index grouped by sectors and industries.
Figure 6.26 FinViz: Standard and Poor’s 500 Index
HOW TO READ IT & WHAT TO LOOK FOR
Look at the high-level groupings to understand the different containing arrangements and establish what the colour association is. Glance across the entire chart to seek out the big, small and medium individual rectangular sizes and perform global comparisons to establish a general ranking of biggest > smallest values. Also identify the largest through to smallest container group of rectangles. If the colour coding is based on quantitative variables, look out for the most eye-catching patterns at the extreme end of the scale(s). If labels are provided (or offered through interactivity), browse around the display looking for categories and values of specific interest. As with any display based on the size of the area of a shape, precise reading of values is hard to achieve and so it is important to understand that treemaps can only aim to provide a
285
single-view gist of the properties of the many components of the whole.
PRESENTATION TIPS
INTERACTIVITY: Typically, a treemap will be presented with interactive features to enable selection/mouseover events to reveal further annotated details and/or drill-down navigation.
ANNOTATION: Group/container labels are often allocated a cell of space but these are not to be read as proportional values. Effective direct value labelling becomes difficult as the rectangles get smaller, so often only the most prominent values might be annotated. Interactive features will generally offer visibility of the relevant labels where possible.
COLOUR: Colour can also be used to provide further categorical grouping distinction if not already assigned to represent a quantitative measure of change.
COMPOSITION: As the tiling algorithm is focused on optimising the dimensions and arrangement of the rectangular shapes, treemaps may not always be able to facilitate much internal sorting of high to low values. However, generally you will find the larger shapes appear in the top left of each container and work outwards towards the smaller constituent parts.
VARIATIONS & ALTERNATIVES
A variation of the treemap sees the rectangular layout replaced by a circular one and the rectangular tiles replaced by organic shapes. These are known as ‘Voronoi treemaps’ as the tiling algorithm is informed by a Voronoi tessellation. The ‘circle packing diagram’, a variation of the ‘bubble chart’, similarly shows many parts to a whole but uses a non- tessellating circular shape/layout. The ‘mosaic plot’ or ‘Marimekko chart’ is similar in appearance to a treemap but, in contrast to the treemap’s hierarchical display, presents a detailed breakdown of quantitative value distributions across several categorical dimensions, essentially formed by varied width stacked bars.
Charts Part-to-whole
286
Venn diagram
ALSO KNOWN AS Set diagram, Euler diagram (wrongly)
REPRESENTATION DESCRIPTION
A Venn diagram shows collections of and relationships between multiple sets. They typically use round or elliptical containers to represent all different ‘membership’ permutations to include all independent and intersecting containers. The size of the contained area is (typically) not important: what is important is in which containing region a value resides, which may be represented through the mark of a text label or ‘point’.
EXAMPLE Comparing sets of permutations for legalities around marijuana usage and same-sex marriage across states of the USA.
Figure 6.27 This Venn Diagram Shows Where You Can Both Smoke Weed and Get a Same-Sex Marriage
287
HOW TO READ IT & WHAT TO LOOK FOR
To read a Venn diagram firstly establish what the different containers are representative of in terms of their membership. Assess the membership of the intersections (firstly ‘all’, then ‘partial’ intersections when involving more than two sets) then work outwards towards the independent container regions where values are part of one set but not part of others. Occasionally there will be a further grouping state outside of the containers that represents values that have no membership with any set at all.
PRESENTATION TIPS
ANNOTATION: Unless you are using point markers to represent membership values, clear labels are vital to indicate how many or which elements hold membership with each possible set combination.
COLOUR: Colour is often used to create more immediate distinction between the intersections and independent parts or members of each
288
container.
COMPOSITION: As the attributes of size and shape of the containers are of no significance there is more flexibility to manipulate the display to fit the number of sets around the constraint of real estate you are facing and to get across the set memberships you are attempting to show. The complexity of creating containers to accommodate all combinations of intersection and independence states increases as the number of sets increases, especially to preserve all possible combinations of intersections between and independencies from all sets. As the number of sets increases, the symmetry of shape reduces and the circular containers are generally replaced with ellipses. While it is theoretically possible to exceed four and five set diagrams, the ability of readers to make sense of the displays diminishes and so they commonly involve only two or three different sets.
VARIATIONS & ALTERNATIVES
A common variation or alternative to the Venn (but often mistakenly called a Venn) is the ‘Euler diagram’. The difference is that an Euler diagram does not need to present all possible intersections with and independencies from all sets. A different approach to visualising sets (especially larger numbers) can be achieved using the ‘UpSet’ technique.
Charts Hierarchies
Dendrogram
ALSO KNOWN AS Node–link diagram, layout tree, cluster tree, tree hierarchy
REPRESENTATION DESCRIPTION
A dendrogram is a node–link diagram that displays the hierarchical
289
relationship across multiple tiers of categorical dimensions. It displays a hierarchy based on multi-generational ‘parent-and-child’ relationships. Starting from a singular origin root node (or ‘parent’) each subsequent set of constituent ‘child’ nodes, a tier below and represented by points, is connected by lines (curved or straight) to indicate the existence of a relationship. Each constituent node may have further sub-constituencies represented in the same way, continuing down through to the lowest tier of detail. Each ‘generational’ tier is presented at the same relative distance from the origin. The layout can be based on either a linear tree structure (typically left to right) or radial tree (outwards from the centre).
EXAMPLE Showing a breakdown of the 200+ beer brands belonging to SAB InBev across different countries grouped by continent.
Figure 6.28 The 200+ Beer Brands of SAB InBev
HOW TO READ IT & WHAT TO LOOK FOR
290
Reading a dendrogram will generally be a highly individual experience based on your familiarity with the subject and your interest in exploring certain hierarchical pathways. The main focus of attention will likely be to find the main clusters from where most constituent parts branch out and to contrast these with the thinner, lighter paths comprising fewer parts. Work left to right (linear) or in to out (radial) through the different routes that stoke your curiosity.
PRESENTATION TIPS
ANNOTATION: With labelling required for each node, depending on the number of tiers and the amount of nodes, the size of the text will need to be carefully considered to ensure readability and minimise the effect of clutter.
COLOUR: Colour would be an optional choice for accentuating certain nodes or applying some further visual categorisation.
COMPOSITION: There are several different layout options to display tree hierarchies like the dendrogram. The common choice is a cluster layout based on the ‘Reingold–Tilford’ tree algorithms that offers a tidying and optimisation treatment for the efficiency of the arrangement of the nodes and connections. The sequencing of sub-constituencies under each node could be logically arranged in some more meaningful way than just alphabetical, though the cataloguing nature of A–Z may suit your purpose. The choice of a linear or radial tree structure will be informed largely by the space you have to work in as well as by the cyclical or otherwise nature of the content in your data. The main issue is likely to be one of legibility if and when you have numerous layers of divisions and many constituent parts to show in a single view.
VARIATIONS & ALTERNATIVES
More advanced applications of dendrograms are used to present hierarchical clustering (in fields such as computational biology) and apply more quantitative meaning to the length of the links and the positioning of the nodes. The ‘tree hierarchy diagram’ offers a similar tree structure but introduces quantitative attributes to the nodes using area marks, such as circles, sized according to a quantitative value. An alternative approach to the dendrogram could involve a ‘linear bracket’. This might show hierarchical structures for data-related sporting competitions with knock-out format. The outer nodes would be the starting point representing all the participating competitors/teams. Each subsequent tier would represent those participants who progressed to the next round, continuing through to the finalists and eventual victors.
291
Charts Hierarchies
Sunburst
ALSO KNOWN AS Adjacency diagram, icicle chart, multi-level pie chart
EXAMPLE Showing a breakdown of the types of companies responsible for extracting different volumes of carbon-based fuels through various activities.
REPRESENTATION DESCRIPTION
A sunburst chart is an adjacency diagram that displays the hierarchical and part-to-whole relationships across multiple tiers of categorical dimensions. In contrast to the dendrogram, the sunburst uses layers of concentric rings, one layer for each generational tier. Each ring layer is divided into parts based on the constituent categorical dimensions at that tier. Each part is represented by a different circular arc section that is sized (in length; width is constant) according to the relative proportion. Starting from the centre ‘parent’ tier, the outward adjacency of the constituent parts of each tier represents the ‘parent-and-child’ hierarchical composition.
Figure 6.29 Which Fossil Fuel Companies are Most Responsible for Climate Change?
292
HOW TO READ IT & WHAT TO LOOK FOR
Reading a sunburst chart will be a highly individual experience based on your familiarity with the subject and your interest in exploring certain hierarchical pathways. The main focus of attention will likely be to find the largest arc lengths, representing the largest single constituent parts, and those layers or tiers with the most constituent parts. Work from the centre outwards through the different routes that stoke your curiosity. Depending on the deployment of colour, this may help you identify certain additional categorical patterns.
PRESENTATION TIPS
INTERACTIVITY: Often interactive mouseover/selection events are the only way to reveal the annotations here.
ANNOTATION: Labelling can be quite difficult to fit into the narrow spaces afforded by small proportion ‘parts’. If interactivity is not an
293
option you may decide to label only those parts that can accommodate the text space.
COLOUR: Colours are often used to achieve further categorical distinction.
COMPOSITION: Sometimes the parent–child (and other generational) relationships could be legitimately reversed, so decisions need to be made about the best hierarchy sequencing to suit the curiosities of the audience. The sequencing of sub-constituencies under each node could also be logically arranged in a meaningful way, more so than just alphabetical, unless the cataloguing nature of A–Z ordering suits your purpose.
VARIATIONS & ALTERNATIVES
Where the sunburst chart uses a radial layout, the ‘icicle chart‘ uses a vertical, linear layout starting from the top and moving downwards. The choice of a linear or radial tree structure will be informed largely by the space you have to work in as well as by the legitimacy of the cyclical nature of the content in your data. A variation on the sunburst chart would be the ‘ring bracket’. This might show a reverse journey for hierarchical data based on something like sporting competitions with knock-out formats. The outer concentric partitions would represent the participant competitors/teams at the start of the process. The length of these arc line parts would be equally distributed across all constituent parts with each subsequent tier representing ‘participants’ who progress forward to the next ‘round’, continuing through to the finalists and eventual victors in the centre.
Charts Correlations
Scatter plot chart
ALSO KNOWN AS Scatter graph
294
REPRESENTATION DESCRIPTION
A scatter plot displays the relationship between two quantitative measures for different categories. Scatter plots are used to explore visually the potential existence, extent or absence of a significant relationship between the plotted variables. The display is formed by points (usually a dot or circle), representing each category and plotted positionally along quantitative x- and y-axes. Sometimes colour is used to distinguish categorical dimensions across all the points. Scatter plots do not work too well if one or both of the quantitative measures has limited variation in value as this especially causes problems of ‘occlusion’, whereby multiple instances of the similar values are plotted on top of each other and essentially hidden from the reader.
EXAMPLE Exploring the relationship between life expectancy and the percentage of healthy years across all countries.
Figure 6.30 How Long Will We Live — And How Well?
HOW TO READ IT & WHAT TO LOOK FOR
Learn what each quantitative axis relates to and make a note of the range of values in each case (min to max). Look at what category or observation each plotted value on the chart refers to and look up any colour associations being used for categorical distinction. Scan the chart looking for the existence of any diagonal trends that might suggest a
295
linear correlation between the variables, or note the complete absence of any pattern, to mean no correlation. Annotations will often assist in determining the significance of any patterns like this. Identify any clusters of points and also look at the gaps, which can be just as revealing. Some of the most interesting observations come from individual outliers standing out separately from others. Look out for any patterns formed by points with similar categorical colour. One approach to reading the ‘meaning’ of the plotted positions involves trying to break down the chart area into a 2 × 2 grid translating what marks positioned in those general areas might mean – which corner is ‘good’ or ‘bad’ to be located in? Remember that ruling out significant relationships can be just as useful as ruling them in.
PRESENTATION TIPS
ANNOTATION: Gridlines can be useful to help make the value estimates clearer and reference lines (such as a trend line of best fit) might aid interpretation. It is usually hard to make direct labelling of all values work well. Firstly, it can be tricky making it clear which value relates to which point, especially when several points may be clustered together. Secondly, it creates a lot of visual clutter. Labelling choices should be based on values that are of most interest to include editorially unless interactive features enable annotations to be revealed through selection or mouseover events. If possible, you might consider putting a number inside the marker to indicate a count of the number of points at the same position if this occurs.
COLOUR: If colours are being used to distinguish the different categories, ensure these are as visibly different as possible. On the occasion where multiple values may be plotted close to or on top of each other, you might need to use semi-transparency to enable overlapping of points to build up a recognisably darker colour compared to other points, indicating an underlying stack of values at the same location on the chart.
COMPOSITION: As the encoding of the plotted point values is based on position along an axis, it is not necessary to start the axes from a zero baseline, so just make the scale ranges as representative as possible of the range of values being plotted. Ideally a scatter plot will have a 1:1 aspect ratio (equally as tall as it is wide), creating a squared area to help patterns surface more evidently. If one quantitative variable (e.g. weight) is likely to be affected by the other variable (e.g. height), it is general practice to place the former on the y-axis and the latter on the x- axis. If you have to use a logarithmic quantitative scale on either or both
296
axes, you need to make this clear to readers so they avoid making incorrect conclusions from the resulting patterns (that might imply correlation if the values were linear, for example).
VARIATIONS & ALTERNATIVES
A ‘ternary plot’ is a variation of the scatter plot through the inclusion of a third quantitative variable axis. The ‘bubble plot’ also incorporates a third quantitative variable, this time through encoding the size of a geometric shape (replacing the point marker). A ‘scatter plot matrix’ involves a single view of multiple scatter plots presenting different combinations of plotted quantitative variables, used to explore possible relationships among larger multivariate datasets. A ‘connected scatter plot’ compares the shifting state of two quantitative measures over time.
Charts Correlations
Bubble plot
ALSO KNOWN AS Bubble chart
REPRESENTATION DESCRIPTION
A bubble plot displays the relationship between three quantitative measures for different categories. Bubble plots are used visually to explore the potential existence, extent or absence of a significant relationship between the plotted variables. In contrast to the scatter plot, the bubble plot plots proportionally sized circular areas, for each category, across two quantitative axes with the size representing a third quantitative measure. Sometimes colour is used to distinguish categorical dimensions across all the shapes.
EXAMPLE Exploring the relationship between rates of murders, burglaries (per 100,000 population) and population across states of the USA.
297
Figure 6.31 Crime Rates by State
HOW TO READ IT & WHAT TO LOOK FOR
Learn what each quantitative axis relates to and make a note of the range of values in each case (min to max). Look at what category or observation each plotted value on the chart refers to. Establish the quantitative size associations for the bubble areas and look up any colour associations being used for categorical distinction. Scan the chart looking for the existence of any diagonal trends that might suggest a linear correlation between the variables, or note the complete absence of any pattern, to mean no correlation. Annotations will often assist in determining the significance of any patterns like this. Identify any clusters of points and also look at the gaps, which can be just as revealing. Some of the most interesting observations come from individual outliers standing out separately from others. Look out for any patterns formed by points with similar categorical colour. What can you learn about the distribution of small, medium or large circles: are they clustered together in similar regions of the chart or quite randomly scattered? One approach to reading the ‘meaning’ of the plotted positions involves trying to break down the chart area into a 2 × 2 grid translating what marks positioned in those general areas might mean – which corner is ‘good’ or ‘bad’ to be located in? Remember that ruling out significant relationships can be just as useful as ruling them in.
298
Estimating and comparing the size of areas is not as easy as it is for judging bar length or dot position. This means that the use of this chart type will primarily be about facilitating a gist – a general sense of the hierarchy of the largest and smallest values.
PRESENTATION TIPS
ANNOTATION: Gridlines can be useful to help make the value estimates clearer and reference lines (such as a trend line of best fit) might aid interpretation. It is usually hard to make direct labelling of all values work well. Firstly, it can be tricky making it clear which value relates to which point, especially when several points may be clustered together. Secondly, it creates a lot of visual clutter. Labelling choices should be based on values that are of most interest to include editorially unless interactive features enable annotations to be revealed through selection or mouseover events.
COLOUR: If colours are being used to distinguish the different categories, ensure these are as visibly different as possible. When a circle has a large value its size will often overlap in spatial terms with other values. The use of outline borders and semi-transparent colours helps with the task of avoiding occlusion (visually hiding values behind others).
COMPOSITION: As the encoding of the plotted area marker values is based on position along an axis, it is not necessary to start the axes from a zero baseline – just make the scale ranges as representative as possible of the range of values being plotted. Make sensible decisions about how large to make the maximum bubble size; this will usually require trial and error experimentation to find the right balance. Ideally a bubble plot will have a 1:1 aspect ratio (equally as tall as it is wide), creating a squared area to help patterns surface more evidently. If one quantitative variable (e.g. weight) is likely to be affected by the other variable (e.g. height), it is general practice to place the former on the y-axis and the latter on the x-axis. Geometric accuracy of the circle size calculations is paramount, since mistakes are often made with circle size calculations: it is the area you are modifying, not the diameter/radius. If you wish to make your bubbles appear as 3D spheres you are essentially no longer representing quantitative values through the size of a geometric area mark, rather the mark will be a ‘form’ and so the size calculation will be based on volume, not area.
VARIATIONS & ALTERNATIVES
If the third quantitative variable is removed, the display would just
299
become a ‘scatter plot’. Variations on the bubble plot might see the use of different geometric areas as the markers, maybe introducing extra meaning from the underlying data through the shape, size and dimensions used.
Charts Correlations
Parallel coordinates
ALSO KNOWN AS Parallel sets
REPRESENTATION DESCRIPTION
Parallel coordinates display multiple quantitative measures for different categories in a single display. They are used visually to explore the relationships and characteristics of multi-dimensional, multivariate data. Parallel coordinates are based on a series of parallel axes representing different quantitative measures with independent axis scales. The quantitative values for each measure are plotted and then connected to form a single line. Each connected line represents a different category record. Colour may be used to differentiate further categorical dimensions. As more data is added the collective ’shape’ of the data emerges and helps to inform the possibility of relationships existing among the different measures. Parallel coordinates look quite overwhelming but remember that they are almost always only used to assist in exploratory work of large and varied datasets, more so than being used for explanatory presentations of data. Generally the greater the number of measures, the more difficult the task of making sense of the underlying patterns will be, so be discerning in your choice of which variables to include. This method does not work for showing categorical (nominal) measures nor does it really offer value with the inclusion of low-range, discrete quantitative variables used (e.g. number of legs per human). Patterns will mean very little when intersecting with such axes (they may be better deployed as a filtering parameter or a coloured categorical separator).
300
EXAMPLE Exploring the relationship between nutrient contents for 14 different attributes across 1,153 different items of food.
Figure 6.32 Nutrient Contents — Parallel Coordinates
HOW TO READ IT & WHAT TO LOOK FOR
Look around the chart and acquaint yourself with what each quantitative measure axis represents. Also note what kind of sequencing of measure has been used: are neighbouring measures significantly paired? Note the range of values along each independent axis so you understand what positions along the scales represent and can determine what higher and lower positions mean. If colour has been used to group related records then identify what these represent. Scan the overall mass of lines to identify any major patterns. Study the patterns in the space between each pair of adjacent axes. This is where you will really see the potential presence or absence of, and nature of, relationships between measures. The main patterns to identify involve the presence of parallel lines (showing consistent relationships), lines converging in similar directions (some correlation) and then complete criss-crossing (negative relationship). Look out for any associations in the patterns across colour groupings. Remember that ruling out significant relationships can be just as useful as ruling them in.
PRESENTATION TIPS
INTERACTIVITY: Parallel coordinates are particularly useful when offered with interactive features, such as filtering techniques, enabling the user to interrogate and manipulate the display to facilitate visual exploration. Additionally, the option to rearrange the sequence of the measures can be especially useful.
301
ANNOTATION: The inclusion of visible annotated features like axis lines, tick marks, gridlines and value labels can naturally aid the readability of the data but be aware of the impact of clutter.
COLOUR: When you are plotting large quantities of records, inevitably there will be over-plotting and this might disguise the real weight of values, so the variation in the darkness of colour can be used to establish density of observations.
COMPOSITION: The ordering of the quantitative variables has to be of optimum significance as the connections between adjacent axes will offer the main way of seeing the local relationships: the patterns will change for every different ordering permutation. Remember that the line directions connecting records are often inconsequential in their meaning unless neighbouring measures have a common scale and similar meaning: the connections are more about establishing commonality of pattern across records, rather than there being anything too significant behind the absolute slope direction/length.
VARIATIONS & ALTERNATIVES
The ‘radar chart’ has similarities with parallel coordinates in that they include several independent quantitative measures in the same chart but on a radial layout and usually only showing data for one record in the same display. A variation on the parallel coordinate would be the ‘Sankey diagram’, which displays categorical composition and quantitative flows between different categorical dimensions or ‘stages’.
Charts Correlations
Heat map
ALSO KNOWN AS Matrix chart, mosaic plot
REPRESENTATION DESCRIPTION
302
A heat map displays quantitative values at the intersection between two categorical dimensions. The chart comprises two categorical axes with each possible value presented across the row and column headers of a table layout. Each corresponding cell is then colour-coded to represent a quantitative value for each combination of category pairing. It is not easy for the eye to determine the exact quantitative values represented by the colours, even if there is a colour scale provided, so heat maps mainly facilitate a gist of the order of magnitude.
EXAMPLE Exploring the connections between different Avengers characters appearing in the same Marvel comic book titles between 1963 and 2015.
Figure 6.33 How the ‘Avengers’ Line-up Has Changed Over the Years
HOW TO READ IT & WHAT TO LOOK FOR
Learn what each categorical dimension relates to and make a note of the range of values in each case, paying attention to the significance of any ordering. Establish the quantitative value associations for the colour scales, usually found via a legend. Glance across the entire chart to locate the big, small and medium shades (generally darker = larger) and perform global comparisons to establish the high-level ranking of biggest > smallest. Scan across each row and/or column to see if there are specific patterns associated with either set of categories. Identify any noticeable exceptions and/or outliers. Perform local comparisons between neighbouring cell’s areas, to identify larger than and smaller
303
than relationships and estimate the relative proportions. Estimate (or read, if labels are present) the absolute values of specific colour scales of interest.
PRESENTATION TIPS
ANNOTATION: Direct value labelling is possible, otherwise a clear legend to indicate colour associations will suffice.
COLOUR: Sometimes multiple different colour hues may be used to subdivide the quantitative values into further distinct categorical groups. Decisions about how many colour-scale levels and what intervals each relates to in value ranges will affect the patterns that emerge. There is no single right answer – you will arrive at it largely through trial and error/experimentation – but it is important to consider, especially when you have a diverse distribution of values.
COMPOSITION: Logical sorting (and maybe even sub-grouping) of the categorical values along each axis will aid readability and may help surface key relationships.
VARIATIONS & ALTERNATIVES
A ‘radial heat map’ offers a structure variation whereby the table may be portrayed using a circular layout. As with any radial display this is only really of value if the cyclical ordering means something for the subject matter. A variation would see colour shading replaced by a measure of pattern density, using a scale of ‘packedness’ to indicate increasing quantitative values. An alternative approach would be the ‘matrix chart’ using size of a shape to indicate the quantitative or a range of point marker to display categorical characteristics.
Charts Connections
Matrix chart
304
ALSO KNOWN AS Table chart
REPRESENTATION DESCRIPTION
A matrix chart displays quantitative values at the intersection between two categorical dimensions. The chart comprises two categorical axes with each possible value presented across the row and column headers of a table layout. Each corresponding cell is then marked by a geometric shape with its area sized to represent a quantitative value and colour often used visually to distinguish a further categorical dimension. While they are most commonly seen using circles, you can use other proportionally sized shapes.
EXAMPLE Exploring the perceived difficulty of fixtures across the season for teams in the premier league 2013–14.
Figure 6.34 Interactive Fixture Molecules
HOW TO READ IT & WHAT TO LOOK FOR
Learn what each categorical dimension relates to and make a note of the range of values in each case, paying attention to the significance of any ordering. Establish the quantitative size associations for the area marks and look up any colour associations being used, both usually found via a legend. Glance across the entire chart to locate the big, small and medium areas and perform global comparisons to establish the high- level ranking of biggest > smallest. Scan across each row and/or column to see if there are specific patterns associated with either set of categories. Identify any noticeable exceptions and/or outliers. Perform local comparisons between neighbouring circular areas, to identify larger than and smaller than relationships and estimate the relative proportions. Estimate (or read, if labels are present) the absolute values
305
of specific geometric areas of interest.
PRESENTATION TIPS
ANNOTATION: Direct value labelling is possible, otherwise be sure to include a clear size legend. Normally this will be more than sufficient as the reader may simply be looking to get a gist of the order of magnitude.
COLOUR: If colours are being used to distinguish the different categories, ensure these are as visibly different as possible.
COMPOSITION: If there are large outlier values there may be occasions when the size of a few circles outgrows the cell it occupies. You might editorially decide to allow this, as the striking shape may create a certain impact, otherwise you will need to limit the largest quantitative value to be represented by the maximum space available within the table’s cell layout. Logical sorting (and maybe even sub- grouping) of the categorical values along each axis will aid readability and may help surface key relationships. The geometric accuracy of the circle size calculations is paramount. Mistakes are often made with circle size calculations: it is the area you are modifying, not the diameter/radius.
VARIATIONS & ALTERNATIVES
A variation may be to remove the quantitative attribute of the area marker, replacing it with a point marker to represent a categorical status to indicate simply a yes/no observation through the presence/absence of a point or through the quantity of points to represent a total. An application of this might be in calendar form whereby a marker in a date cell indicates an instance of something. It could also employ a broader range of different categorical options; in practice any kind of marker (symbol, colour, photograph) could be used to show a characteristic of the relationship at each coordinate cell. An alternative might be the ‘heat map’ which colour-codes the respective cells to indicate a relationship based on a quantitative measure.
Charts Connections
306
Node–link diagram
ALSO KNOWN AS Network diagram, graph, hairballs
REPRESENTATION DESCRIPTION
Node–link diagrams display relationships through the connections between categorical ‘entities’. The entry-level version of this type of diagram displays entities as nodes (represented by point marks and usually including a label) with links or edges (represented by lines) depicting the existence of connections. The connecting lines will often display an attribute of direction to indicate the influencer relationship. In some versions a quantitative weighting is applied to the show relationship strength, maybe through increased line width. Replacing point marks with a geometric shape and using attributes of size and colour is a further variation. Often the complexity seen in these displays is merely a reflection of the underlying complexity of the subject and/or system upon which the data is based, so oversimplifying can compromise the essence of such content.
EXAMPLE Exploring the connections of voting patterns for Democrats and Republicans across all members of the US House of Representatives from 1949 to 2012.
Figure 6.35 The Rise of Partisanship and Super-cooperators in the U.S.
307
HOW TO READ IT & WHAT TO LOOK FOR
The first thing to consider is what entity each node (point or circular area) represents and what the links mean in relationship terms. There may be several other properties to acquaint yourself with, including attributes like the size of the node areas, the categorical nature of colouring, and the width and direction of the connections. Across the graph you will mainly be seeking out the clusters that show the nodes with the most relationships (representative of influencers or hubs) and those without (including outliers). Small networks will generally enable you to look closely at specific nodes and connections and easily see the emerging relationships. When datasets are especially large, consisting of thousands of nodes and greater numbers of mutual connections, the displays can seem overwhelmingly cluttered and will be too dense to make many detailed observations at node–link level. Instead, just relax and know that your readability will be about a higher level sense- making of the clusters/hubs and main outliers.
PRESENTATION TIPS
INTERACTIVITY: Node–link diagrams are particularly useful when offered with interactive features, enabling the user to interrogate and manipulate the display to facilitate visual exploration. The option to apply filters to reduce the busy-ness of the visual and enable isolation of individual node connections helps users to focus on specific parts of the network of interest.
308
ANNOTATION: The extent of annotated features tends to be through the inclusion of value labels for each entity. Accommodating the relative word sizes on each node can be difficult to achieve with real elegance (once again that is where interactivity adds value, through the select/mouseover event to reveal the label).
COLOUR: Aside from the possible categorical colouring of each node, decisions need to be made about the colour of the connecting lines, especially on deciding how prominent these links will be in contrast to the nodes.
COMPOSITION: Composition decisions are where most of the presentation customisation exists. There are several common algorithmic treatments used to compute custom arrangements to optimise network displays, such as force-directed layouts (using the physics of repulsion and springs to amplify relationships) and simplifying techniques (such as edge bundling to aggregate/summarise multiple similar links).
VARIATIONS & ALTERNATIVES
There are many derivatives of the node–link diagram, as explained, based on variations in the use of different attributes. ‘Hive plots’ and ‘BioFabric’ offer alternative approaches based on replacing nodes with vertices.
Charts Connections
Chord diagram
ALSO KNOWN AS Radial network diagram, arc diagram (wrongly)
REPRESENTATION DESCRIPTION
A chord diagram displays relationships through the connections between and within categories. They are formed around a radial display
309
with different categories located around the edge: either as individual nodes or proportionally sized segments (arcs) of the circumference according to a part-to-whole breakdown. Emerging inwards from each origin position are curved lines that join with other related categorical locations around the edge. The connecting lines are normally proportionally sized according to a quantitative measure and a directional or influencing relationship is often indicated. The perceived readability of the chord diagram will always be influenced by the quantity and range of values being plotted. Small networks will enable a reader to look closely at specific categories and their connections to see the emerging relationships easily; larger systems will look busy through the network of lines but they can still provide windows into complex networks of influence. Often the complexity seen in these displays is merely a reflection of the underlying complexity of the subject and/or system upon which the data is based, so oversimplifying can compromise the essence of such content.
EXAMPLE Exploring the connections of migration between and within 10 world regions based on estimates across five-year intervals between 1990 and 2010.
Figure 6.36 The Global Flow of People
310
HOW TO READ IT & WHAT TO LOOK FOR
First determine how categories are displayed around the circumference, either as nodes or part-to-whole arcs, and identify each one individually. Consider the implication of the radial sorting of these categorical values and, if based on part-to-whole sizes, establish a sense of the largest > smallest arc lengths. Colour-coding may be applied to the categories so note any associations. Look inside the display to determine what relationships the connecting lines represent and check for any directional significance. Look closer at the tangled collection of lines criss-crossing this space, noting the big values (usually through line weight or width) and the small ones. Avoid being distracted by the distance a line travels, which is just a by-product of the outer categorical arrangement: a long connecting line is just as significant a relationship as a short one. For this reason, pay close attention to any connecting lines that have very short looping distances to adjacent categories. Are there any patterns of lines heading towards or leaving certain categories?
311
PRESENTATION TIPS
INTERACTIVITY: Chord diagrams are particularly useful when offered with interactive features, enabling the user to interrogate and manipulate the display to facilitate visual exploration. The option to apply filters to reduce the busy-ness of the visual and enable isolation of individual node connections helps users to focus on specific parts of the network of interest.
ANNOTATION: Annotated features tend to be limited to value labelling of the categories around the circumference and, occasionally, directly onto the base or ends of the connecting lines (usually just those that are large enough to accommodate them).
COLOUR: Aside from the categorical colouring of each node, decisions need to be made about the colour of the connecting lines, especially on deciding how prominent these links will be in contrast to the nodes. Sometimes the connections will match the origin or destination colours, or they will combine the two (with a start and end colour to match the relationship).
COMPOSITION: The main arrangement decisions come through sorting, firstly by generating as much logical meaning from the categorical values around the edge of the circle and secondly by deciding on the sorting of the connecting lines in the z-dimension – if many lines are crossing, there is a need to think about which will be on top and which will be below. Showing the direction of connections can be difficult as there is so little room for manoeuvring many more visual attributes, such as arrows or colour changes. One common, subtle solution is to pull the destination join back a bit, leaving a small gap between the connecting line and the destination arc. This then contrasts with connecting lines that emerge directly from the categorical arcs, showing it is their origin.
VARIATIONS & ALTERNATIVES
The main alternatives would be to consider variations of the ‘node–link diagram’ or, specifically, the ‘arc diagram’, which offers a further variation on the theme of networked displays, placing all the nodes along a baseline and forming connections using semi-circular arcs, rather than using a graph or radial layout.
Charts Connections
312
Sankey diagram
ALSO KNOWN AS Alluvial diagram
REPRESENTATION DESCRIPTION
Sankey diagrams display categorical composition and quantitative flows between different categorical dimensions or ‘stages’. The most common contemporary form involves a two-sided display, with each side representing different (but related) categorical dimensions or different states of the same dimension (such as ‘before and after’). On each side there is effectively a stacked bar chart displaying proportionally sized and differently coloured (or spaced apart) constituent parts of a whole. Curved bands link each side of the display to represent connecting categories (origin and destination) with the proportionally sized band (its thickness) indicating the quantitative nature of this relationship. Some variations involve multiple stages and might present attrition through the diminution size of subsequent stacks. Traditionally the Sankey has been used as a flow diagram to visualise energy or material usage across engineering processes. It is closely related to the ‘alluvial diagram’, which tends to show changes in composition and flow over time, but the Sankey label is often applied to these displays also.
EXAMPLE Exploring the seat changes among political parties between the 2010 and 2015 UK General Elections.
Figure 6.37 UK Election Results by Political Party, 2010 vs 2015
313
HOW TO READ IT & WHAT TO LOOK FOR
Based on the basic two-sided version of the Sankey diagram, look down both sides of the chart to learn what states are represented and what the constituent categories are. Pay close attention to the categorical sorting and pick out the large and small values on each side. Then look at the connecting lines, making observations about the largest and narrowest bands and noting any that seem to be mostly redistributed into a different category compared to those that just join with the same. Notice any small break-off bands that seem to cross the height of the whole chart, perhaps representing a more dramatic change or diversion between states. As with most network-type visualisations, the perceived readability of the Sankey diagram will always be influenced by the quantity and range of values being plotted, as well as the number of different states presented.
PRESENTATION TIPS
INTERACTIVITY: Sankey diagrams are particularly useful when offered with interactive features, enabling the user to interrogate and manipulate the display to facilitate visual exploration. The option to apply filters to reduce the busy-ness of the visual and enable isolation of
314
individual node connections helps users to focus on specific parts of the network of interest.
ANNOTATION: Annotated features tend to be limited to value labelling of the categories that make up each ‘state’ stack.
COLOUR: Colouring is often used visually to indicate the categories of the connecting bands, though it can get a little complicated when trying to combine a sense of change through an origin category colour blending with a destination category colour when there has been a switch.
COMPOSITION: The main arrangement decisions come through sorting, firstly by generating as much logical meaning from the categorical values within the stacks and, secondly, by deciding on the sorting of the connecting lines in the z-dimension – if many lines are crossing, there is a need to think about which will be on top and which will be below. There is no significant difference between a landscape or portrait layout, which will depend on the subject matter ‘fit’ and the space within which you have to work. Try to ensure that the sorting of the categorical dimensions is as logical and meaningful as possible.
VARIATIONS & ALTERNATIVES
The concept of a Sankey diagram showing composition and flow can also be mapped onto a geographical projection as one of the variations of the ‘flow map’. You could use a ‘chord diagram’ as an alternative to show how larger networks are composed proportionally and in their connections. Showing how component parts have changed over time could just be displayed using a ‘stacked area chart’. A ‘funnel chart’ is a much simplified display to show how a single value changes (usually diminishing) across states, for topics like sales conversion. This often is based on a funnel-like shape formed by a wide bar at the top (those entering the system) and then gradually narrower bars, stage by stage towards the end state.
Charts Trends
315
Line chart
ALSO KNOWN AS Fever chart, stock chart
REPRESENTATION DESCRIPTION
A line chart shows how quantitative values for different categories have changed over time. They are typically structured around a temporal x- axis with equal intervals from the earliest to latest point in time. Quantitative values are plotted using joined-up lines that effectively connect consecutive points positioned along a y-axis. The resulting slopes formed between the two ends of each line provide an indication of the local trends between points in time. As this sequence is extended to plot all values across the time frame it forms an overall line representative of the quantitative change over time story for a single categorical value. Multiple categories can be displayed in the same view, each represented by a unique line. Sometimes a point (circle/dot) is also used to substantiate the visibility of individual values. The lines used in a line chart will generally be straight. However, sometimes curved line interpolation may be used as a method of estimating values between known data points. This approach can be useful to help emphasise a general trend. While this might slightly compromise the visual accuracy of discrete values if you already have approximations, this will have less impact.
EXAMPLE Showing changes in percentage income growth for the Top 1% and Bottom 90% of earners in the USA between 1917 and 2012.
Figure 6.38 The Fall and Rise of U.S. Inequality, in 2 Graphs
316
HOW TO READ IT & WHAT TO LOOK FOR
Firstly, learn about the axes: what is the time period range presented on the x-axis (and in what order) and what is the range of quantitative values shown on the y-axis, paying particular attention to the origin value (which may not be zero)? Inside the chart, determine what categories each line represents: for single lines this will usually be clear from the chart title, for multiple lines you might have direct labelling or a legend to learn colour associations. Think about what high and low values mean: is it ‘good’ to be large/small, increasing or decreasing? Glance at the general patterns (especially if there are many) looking for observations such as any trends (short or long term), any sudden moments of a rise or fall (V- or W -shapes, or inverted), any sense of seasonal or cyclical patterns, any points of interest where lines cross each other or key thresholds that are reached/exceeded. Can you mentally extrapolate from the values shown any sense of a forecasted trend? Avoid jumping to spurious interpretations if you see two line series following a similar pattern; this does not necessarily mean that one thing has caused the other, it might just be coincidence. Then look more closely at categories of interest and at patterns around specific moments in time, and pick out the peak, low, earliest and latest values for each line. Where available, compare the changing quantities against annotated references such as targets, forecast, previous time periods, range bands, etc.
PRESENTATION TIPS
317
INTERACTIVITY: Interactivity may be especially helpful if you have many categories and wish to enable the user to isolate (in focus terms) a certain line category of interest.
ANNOTATION: Chart apparatus devices like tick marks and gridlines in particular can be helpful to increase the accuracy of the reading of the quantitative values. If you have axis labels you should not need direct labels on each value point – this will be label overload. You might choose to annotate specific values of interest (highest, lowest, specific milestones). Think carefully about what is the most useful and meaningful interval for your time axis labelling. When several categories are being shown, if possible, try directly to label the categories shown by each line, maybe at the start or end position.
COLOUR: When many categories are shown it may be that only certain emphasised lines of interest possess a colour and a label – the rest are left in greyscale for context.
COMPOSITION: Composition choices are mostly concerned with the chart’s dimensions: its aspect ratio, how high and wide to make it. The sequencing of values tends to be left to right for the sequence of the time-based x-axis and low rising to high values on the y-axis; you will need a good (and clearly annotated) reason to break this convention. Line charts do not always need the y-axis to start at zero, as we are not judging the size of a bar, rather the position along an axis. You should expect to see a zero baseline if zero has some critical significance in the interpretation of the trends. If your y-axis origin is not going to be zero, you might include a small gap between the x-axis and the minimum so that it is not implied. Be aware that the upward and downward trends on a line chart can seem more significant if the chart width is narrow and less significant if it is more stretched out. There is no single rule to follow here but a useful notion involves ‘banking to 45°’ whereby the average slope angle across your chart heads towards 45°. While it is impractical to actually measure this, judging by eye tends to be more than sufficient.
VARIATIONS & ALTERNATIVES
Variations of the line chart may include the ‘cumulative line chart’ or ‘step chart’. ‘Spark lines’ are mini line charts that aim to occupy almost only a word’s length amount of space. Often seen in dashboards where space is at a premium and there is a desire to optimise the density of the display. ‘Bar charts’ can also be used to show how values look over time when there is perhaps greater volatility in the quantitative values across the time period and when the focus is on the absolute values at
318
each point in time, more so than trends. Sometimes a line chart can show quantitative trends over continuous space rather than time. For showing ranking over time, consider the ‘bump chart’, and for before and after comparisons, the ‘slope graph’.
Charts Trends
Bump chart
ALSO KNOWN AS
REPRESENTATION DESCRIPTION
A bump chart shows how quantitative rankings for categories have changed over time. They are typically structured around a temporal x- axis with equal intervals from the earliest to latest point in time. Quantitative rankings are plotted using joined-up lines that effectively connect consecutive points positioned along a y-axis (typically top = first). The resulting slopes formed between the two ends of each line provide an indication of the local ranking trends between points in time. As this sequence is extended to plot all values across the time frame it forms an overall line representative of the ranking story for a single categorical value. Multiple categories are often displayed in the same view, showing how rankings have collectively changed over time. Sometimes a point (circle/dot) mark is also used to substantiate the connected visibility of category lines, as is colour (for the lines and/or the points).
EXAMPLE Showing changes in rank of the most populated US cities at each census between 1790 and 1890.
Figure 6.39 Census Bump: Rank of the Most Populous Cities at Each Census, 1790—1890
319
HOW TO READ IT & WHAT TO LOOK FOR
Firstly, you need to learn about the axes. What is the time period range presented on the x-axis (and in what order)? What are the range of quantitative rankings shown on the y-axis (check that the ranks start at 1 from the top downwards)? Inside the chart, determine what categories each line represents: this might be explained through direct labelling, a colour legend, interactivity or through differentiating point marker attributes of colour/shape/pattern. Think about what high and low ranks mean: is it ‘good’ to be high up the rankings and is it better to be moving up or down? Consider the general patterns to look for observations such as consistent trends (largely parallel lines) or completely non-relational patterns (lines moving in all directions). Are there any prominent stories of categories that have had a sudden rise or fall (V- or W-shapes, or inverted)? Is there any evidence of seasonal or cyclical patterns, any key points of interest where lines cross each other or key thresholds that are reached/exceeded? Next, look more closely at categories of interest and at patterns around specific moments in time, and pick out the peak, low, earliest and latest values for each line.
PRESENTATION TIPS
INTERACTIVITY: Interactivity is usually necessary with bump charts, especially if you have many categories and wish to enable the user to isolate (in focus terms) a certain line category of interest.
320
ANNOTATION: The ranking labels can be derived from the vertical position along the scale so direct labelling is usually unnecessary. You might choose to annotate specific values of interest (highest, lowest, specific milestones). Think carefully about what is the most useful and meaningful interval for your time axis labelling.
COLOUR: Often, with many categories to show in the same chart, the big challenge is to distinguish each line, especially as they likely criss- cross often with others. Using colour association can be useful for less than 10 categories, but for more than that you really need to offer the interactivity or maybe decide that only certain emphasised lines of interest will possess a colour and the rest are left in greyscale for context.
COMPOSITION: The sequencing of values tends to be left to right for the sequence of the time-based x-axis with high rankings (low number) on the y-axis moving downwards. You will therefore need a good (and clearly annotated) reason to break this convention.
VARIATIONS & ALTERNATIVES
Alluvial diagrams (similar to Sankey diagrams) can show how rankings have changed over time while also incorporating a component of quantitative magnitude. This approach is effectively merging the ‘bump chart’ with the ‘stacked area chart’. Consider ‘line charts’ and ‘area charts’ if the ranking is of secondary interest to the absolute values.
Charts Trends
Slope graph chart
ALSO KNOWN AS Slope chart
REPRESENTATION DESCRIPTION
A slope graph shows a ‘before and after’ display of changes in
321
quantities for different categories. The display is based on (typically) two parallel quantitative axes with a consistent scale range to cover all possible quantitative values. A line is plotted for each category connecting the two axes together with the vertical position on each axis representing the respective quantitative values. Sometime a dot is also used to further substantiate the visibility of the value positions. These connecting lines form slopes that indicate the upward, downward or stable trend between points in time. The resulting display incorporates absolute values, reveals rank and, of course, shows change between time. Colours are often used visually to distinguish different categorical lines, otherwise this can be used to surface visibly the major trend states (up, down, no change). A slope graph works less well when all values (or the majority) are going in the same direction; consider alternatives if this is the case.
EXAMPLE Showing changes in the share of power sources across all US states between 2004 and 2014.
Figure 6.40 Coal, Gas, Nuclear, Hydro? How Your State Generates Power
HOW TO READ IT & WHAT TO LOOK FOR
322
Firstly, learn about the axes: what are the two points in time being presented and what is the possible range of quantitative values shown on the y-axis, checking that the ranks start from the top down? Inside the chart, learn what each category line relates to and determine what categories each line represents: this might be explained through direct labelling, a colour legend, or through interactivity. Think about what upward, downward and stable trends mean: is it ‘good’ to be moving up or down? Is it more interesting to show no change? Look at the general patterns to observe such things as consistent trends (largely parallel lines in either direction) or completely non-relational patterns (lines moving in all directions). Colour may be used to accentuate the distinction between upward and downward trends. Are there any prominent stories of categories that have had a dramatic rise or fall? Even if no values have dramatically altered, that in itself can be an important finding, especially if change was expected. Next, look more closely at categories of interest and pick out the highest and lowest values on each side to learn about those stories. Look for the gaps where there are no values, and at outlier values too, to see if some sit outside the normal value clusters.
PRESENTATION TIPS
INTERACTIVITY: Depending on the number of category values being presented, slope graphs can become quite busy, especially if there are bunches of similar values and slope transitions. This also causes a problem with accommodating multiple labels on the same value. On these occasions you might find interactive slope graphs to help filter/exclude certain values.
ANNOTATION: Labelling of each category will get busy, especially when there are shared values, so you might choose to annotate specific values of interest (highest, lowest, of editorial interest).
COLOUR: Often when you have many categories to show in the same chart the big challenge is to distinguish each line, especially as they likely criss-cross often with others. Using colour association can be useful for less than 10 categories usually with direct labelling on the left and/or right of the chart.
COMPOSITION: The aspect ratio of the slope graph (height and width) will often be determined by the space you have to work with.
VARIATIONS & ALTERNATIVES
Rather than showing a before and after story, some slope graphs are
323
used to show the relationship between different quantitative measures for linked categories. In this case the connecting line is not indicative of a directional relationship, just the relationship itself. An alternative option would be the ‘connected dot plot’ which can also show before and after stories and is a better option when all values are moving in the same direction.
Charts Trends
Connected scatter plot
ALSO KNOWN AS Trail chart
REPRESENTATION DESCRIPTION
A connected scatter plot displays the relationship between two quantitative measures over time. The display is formed by plotting marks like a dot or circle for each point in time at the respective coordinates along two quantitative x- and y-axes. The collection of individual points is then connected (think of a dot-to-dot drawing puzzle) using lines joining each consecutive point in time to form a sequence of change. Generally there would only be a single connected line plotted on a chart to avoid the great visual complexity of overlaying several in one display. However, if multiple categories are to be included, colour is typically used to distinguish each series.
EXAMPLE Showing changes in the daily price and availability of Super Bowl tickets on the secondary market four weeks prior to the event across five Super Bowl finals.
Figure 6.41 Holdouts Find Cheapest Super Bowl Tickets Late in the Game
324
HOW TO READ IT & WHAT TO LOOK FOR
Learn what each quantitative axis relates to and make a note of the range of values in each case (min to max). Look at what each plotted value on the chart refers to in terms of its date label and determine the meaning of line direction. It usually helps to parse your thinking by considering what higher/lower values mean for each quantitative axis individually and then combining the joint meaning thereafter. Try to follow the chart from the start to the end, mapping out in your mind the sequence of a narrative as the values change in all directions and noting the extreme values in the outer edges of your line’s reach. Look at the overall pattern of the connected line: is it consistently moving in one direction? Does it ebb and flow in all directions? Does it create a spiral shape? Compare consecutive points for a more focused view of change between two points.
PRESENTATION TIPS
INTERACTIVITY: The biggest challenge is making the connections and the sequence as visible as possible. This becomes much harder when values change very little and/or they loop back almost in spiral
325
fashion, crossing back over themselves. It is especially hard to label the sequential time values elegantly. One option to overcome this is through interactivity and particularly through animated sequences which build up the display, connecting one line at a time and unveiling the date labels as time progresses. It is often the case that only one series will be plotted. However, interactive options may allow the user to overlay one or more for comparison, switching them on and off as required.
ANNOTATION: Connected scatter plots are generally seen as one of the most complex chart types for the unfamiliar reader to work out how to read, given the amount of different attributes working together in the display. It is therefore vital that as much help is given to the reader as possible with ‘how to read’ guides and illustrations of what the different directions of change mean.
COLOUR: Colour is only generally used to accentuate certain sections of a sequence that might represent a particularly noteworthy stage of narrative.
COMPOSITION: As the encoding of the plotted point values is based on position along an axis, it is not necessary to start the axes from a zero baseline – just make the scale ranges as representative as possible of the range of values being plotted. Ideally a connected scatter plot will have a 1:1 aspect ratio (equally as tall as it is wide), creating a squared area to help patterns surface more evidently. If one quantitative variable (e.g. weight) is likely to be affected by the other variable (e.g. height), it is general practice to place the former on the y-axis and the latter on the x-axis.
VARIATIONS & ALTERNATIVES
The ‘comet chart’ is to the connected scatter plot what the ‘slope graph’ is to the ‘line chart’ – a summarised view of the changing relationships across two quantitative values between just two points in time. Naturally a reduced variation of the connected scatter plot is simply the ‘scatter plot’ where there is no time dimension or elements of connectedness.
Charts Trends
326
Area chart
ALSO KNOWN AS
REPRESENTATION DESCRIPTION
A line chart shows how quantitative values for different categories have changed over time. They are typically structured around a temporal x- axis with equal intervals from the earliest to latest point in time. Quantitative values are plotted using joined-up lines that effectively connect consecutive points positioned along a y-axis. The resulting slopes formed between the two ends of each line provide an indication of the local trends between points in time. As this sequence is extended to plot all values across the time frame it forms an overall line representative of the quantitative change over time story for a single categorical value. To accentuate the magnitude of the quantitative values and the change through time the area beneath the line is filled with colour. The height of each coloured layer at each point in time reveals its quantity. Area charts can display values for several categories, using stacks, to show also the changing part-to-whole relationship.
EXAMPLE Showing changes in the average monthly price ($ per barrel) of crude oil between 1985 and 2015.
Figure 6.42 Crude Oil Prices (West Texas Intermediate), 1985—2015
327
HOW TO READ IT & WHAT TO LOOK FOR
Firstly, learn about the axes: what is the time period range presented on the x-axis (and in what order) and what is the range of quantitative values shown on the y-axis, paying particular attention to whether it is a percentage or absolute based scale? Inside the chart, determine what categories each area layer represents: for single areas this will usually be clear from the chart title, for multiple areas you might have direct labelling or a nearby legend to learn colour associations. Think about what high and low values mean: is it ‘good’ to be large/small, increasing or decreasing? Glance at the general patterns (especially if there are many layers), looking at the visible ‘thickness’ of the coloured layers. At what points are the values highest or lowest? When are they growing or shrinking as the time axis moves along? If there are multiple categories, which ones take up the largest and smallest slices of the overall total? Are there any trends (short or long term), any sudden moments of a rise or fall, any sense of seasonal or cyclical patterns? If there are multiple categories, look more closely at individual layers of interest.
PRESENTATION TIPS
ANNOTATION: Direct labelling of quantitative values will get far too busy so you might choose to annotate specific values of interest (highest, lowest, specific milestones). Think about the most useful
328
interval for your axis labelling. As ever there is no single rule, so adopt the Goldilocks principle of not too many, not too few. If you have a stacked area chart, try directly to label the category layers shown as closely as possible (if the heights allow it) or at least ensure any colour associations are easily identifiable through a nearby legend. Think carefully about what is the most useful and meaningful interval for your time axis labelling.
COLOUR: If you are using a stacked area chart, ensure the categorical layers have sufficiently different colours so that their distinct reading can be efficiently performed.
COMPOSITION: Similar to the line chart, the area chart’s dimensions should ideally utilise an aspect ratio that optimises the readability through 45° banking (roughly judging the average slope angle). The sequencing of values tends to be left to right for the sequence of the time-based x-axis and low rising to high values on the y-axis; you will need a good (and clearly annotated) reason to break this convention. Unlike the line chart, the quantitative axis for area charts must start at zero as it is the height of the coloured areas under each line that helps readers to perceive the quantitative values. Do not have overlapping categories on the same chart because it makes it very difficult to see (imagine hills behind hills, peaking out and then hiding behind each other). Rather than stacking categories you might consider using small multiples, especially as this will present the respective displays from a common baseline (and make reading sizes a little easier).
VARIATIONS & ALTERNATIVES
Like area charts, ‘alluvial diagrams’ display proportional stacked layers for multiple categories showing the absolute value change over time. However, they also show the evolving ranks, switching the relative ordering of each layer of values based on the current magnitude. Some deployments of the area chart are not plotted over time but over continuous dimensions of space, perhaps showing the changing nature of a given quantitative measure along a given route. When you have many concurrent layers to show and these layers start and stop at different times, a ‘slope graph’ is worth considering.
Charts Trends
329
Horizon chart
ALSO KNOWN AS
EXAMPLE Showing percentage changes in price for selected food items in the USA between 1990 and 2015.
REPRESENTATION DESCRIPTION
Horizon charts show how quantitative values for different categories have changed over time. They are valuable for showing changes over time for multiple categories within space-constrained formats (such as dashboards). They are structured around a series of rows each showing changes in quantitative values for a single category. The temporal x- axis has equal intervals from the earliest to latest point in time. Quantitative values are plotted using joined-up lines that connect consecutive points positioned along a value y-axis. The resulting slopes formed between the ends of each line provide an indication of the local trends between two points in time. As this sequence is extended to plot all values across the time frame it forms an overall line representative of the quantitative changes. To accentuate the magnitude of the quantitative values the area beneath the line is filled with colour. Negative values are highlighted in one colour, positive values in another colour. Variations in colour lightness are used to indicate different degrees or bands of magnitudes, with the extremes getting darker. Negative value areas are then flipped from underneath the baseline to above it, joining the positive values but differentiated in their polarity by colour. Finally, like slicing off layers of a mountain, each distinct threshold band that sits above the imposed maximum y- axis scale is chopped off and dropped down to the baseline, in front of its foundation base. The final effect shows overlapping layers of increasingly darker colour-shaded areas all occupying the same vertical space with combinations of height, colour and shade representing the values.
Figure 6.43 Percentage Change in Price for Select Food Items, Since
330
1990
HOW TO READ IT & WHAT TO LOOK FOR
Firstly, learn about the category rows: what do they represent and in what order are they presented? Next, the chart axes: what is the time period range presented on the x-axis (and in what order) and what is the range of quantitative values shown on the y-axis, paying attention to whether it is a percentage or absolute value scale? Next, what are the colour associations (for positive and negative values) and the different shaded banding thresholds? Think about what high and low values mean: is it ‘good’ to be large/small, increasing or decreasing? Glance at the general patterns over time, looking at the most visible dark areas of each colour polarity: where have values reached a peak in either direction? Maybe then separate your reading between looking at the positive value insights and then the negative ones: which chunks of
331
colour are increasing in value (darker) or shrinking (getting lighter) as the time axis moves along? Where can you see most empty space, indicating low values? Are there any trends (short or long term), any sudden moments of a rise or fall, any sense of seasonal or cyclical patterns, any points of interest where lines cross each other or key thresholds that are reached/exceeded? Then look more closely at categories of interest, assessing their own patterns around specific moments in time and picking out the peak, low, earliest and latest values for each row.
PRESENTATION TIPS
ANNOTATION: The decisions around annotations are largely reduced to labelling the category rows. Such is the busy-ness of the chart areas that any direct labelling is going to clutter the display too much: horizon charts are less about precise value reading and more about getting a sense of the main patterns, so avoid the temptation to over-label. Think carefully about what is the most useful and meaningful interval for your time axis labelling.
COLOUR: Colour decisions mainly concern the choices of quantitative scale bandings to show the positive and negative value ranges.
COMPOSITION: The height of the chart area in which you can accommodate a single row of data will have an influence on the entire construction of the horizon chart. It will often involve an iterative/trial and error process, looking at the range of quantitative values across each category, establishing the most sensible and meaningful thresholds within these range and then fixing the y-axis scales accordingly. Try to ensure the sorting of the main categorical rows is as logical and meaningful as possible.
VARIATIONS & ALTERNATIVES
An alternative to the horizon chart is the entry-level single category ‘area chart’, which does not suffer the same constraints of restrictions to the vertical scale. For space-constrained displays, ‘spark lines’ would offer an option suitable to such situations and easily accommodate multiple category displays.
Charts Trends
332
Stream graph
ALSO KNOWN AS Theme river
REPRESENTATION DESCRIPTION
A stream graph shows how quantitative values for different categories have changed over time. They are generally used when you have many constituent categories at any given point in time and these categories may start and stop at different points in time (rather than continue throughout the presented time frame). As befitting the name, their appearance is characterised by a flowing, organic display of meandering layers. They are typically structured around a temporal x-axis with equal intervals from the earliest to latest point in time. Quantitative values are plotted using joined-up lines that effectively connect consecutive points to quantify the height above a local baseline, which is not a stable zero baseline but rather a shifting shape formed out of other category layers. To accentuate the size of the category’s height at any given point the area beneath the line is filled with colour. The height of each coloured layer at each point in time reveals its quantity. This colour is often used to further represent a quantitative value scale or to associate with categorical colours. The stacking arrangement of the different categorical streams goes above and below the central axis line to optimise the layout but not with any implication of polarity.
EXAMPLE Showing changes in the total domestic gross takings ($US) and the longevity of all movies released between 1986 and 2008.
Figure 6.44 The Ebb and Flow of Movies: Box Office Receipts 1986— 2008
333
HOW TO READ IT & WHAT TO LOOK FOR
Firstly, determine what is the time period presented on the x-axis (and in what order). In most stream graphs you do not see the quantitative y- axis scale because the level of reading is more about getting a gist for the main patterns in a relative sense rather than an absolute one. You might find that the colouring of layers has a quantitative scale or categorical association so look for any keys. Also, you will often find guides to help estimate the quantitative heights of each layer. Think about what high and low values mean: is it ‘good’ to be large/small, increasing or decreasing? Glance at the general patterns over time. Remember that above or below means nothing in the sense of polarity of values, so your focus is on the entirety of the collective shape. Look for the largest peaks and the shallowest troughs, possible seasonal patterns or the significant moments of change. Note where these patterns occur in relation to the timescale. Can you see any prominently tall (big values) or wide (long-duration) layers? Notice when layers start and end, noting times when there are many concurrent categories and when there are few. Pick out the layers of personal interest and assess their patterns over time. Do not spend too much effort trying to estimate precise values of height, but keep your focus on the bigger picture level. It is often useful to rotate the display so the streams are travelling vertically, offering a different perspective and removing the instinct to see positive values above and negative values below the central axis.
PRESENTATION TIPS
334
INTERACTIVITY: If interactivity is a possibility, this could enable selection or mouseover events to reveal annotated values at any given point in time or to filter the view.
ANNOTATION: Chart apparatus devices are generally of limited use in a stream graph with the priority on a general sense of pattern more than precision value reading. Direct labelling of categories is likely to be quite busy but may be required, at least to annotate the most interesting patterns (highest, lowest, specific milestones). Think carefully about what is the most useful and meaningful interval for your time axis labelling.
COLOUR: Ensure any colour associations or size guides are easily identifiable through a nearby legend.
COMPOSITION: Composition choices are firstly concerned with the landscape or portrait layout. This will largely be informed by the format and space of your outputs and the meaning of the data. The stream layers are often smoothed, giving them an aesthetically organic appearance, both individually and collectively. This is achieved via curved line interpolation.
VARIATIONS & ALTERNATIVES
The fewer categorical series you have in your data, the more likely a stacked ‘area chart’ is going to best-fit your needs. You could consider a stacked ‘bar chart’ over time also, but there is less chance of maintaining the connected visibility of continuous categorical series via a singular shape.
Charts Activities
Connected timeline
ALSO KNOWN AS Relationship timeline, storyline visualisations,
335
swim-lane chart
REPRESENTATION DESCRIPTION
A connected timeline displays the duration, milestones and categorical relationships across a range of categorical ‘activities’. It represents a particularly diverse and creative way of showing changes over time and so involves many variations in approach. The structure is generally formed of time-based quantitative x-axis and categorical y-axis lanes. Each categorical activity will commence at a point in time and from within a vertical category ‘family’. Over time, the line will progress, possibly switching to a different categorical lane position as the nature of the activity alters. The lines may be of fixed width or proportionally weighted to represent a quantitative measure. Some activity lines may cease, restart or merge with others to build a multi-faceted narrative. Colour can also be used to present further relevant detail. The main issue with any connected timeline approach is simply the complexity of the content and the number of moving parts crossing over the display. As there are many entry points into reading such a timeline there can be inefficiency in the reading process, but this is usually proportional simply to the subject at hand and you may not wish to see these nuances being removed.
EXAMPLE Showing changes in US major college football programme allegiance to different conferences between 1965 and 2015.
Figure 6.45 Tracing the History of N.C.A.A. Conferences
336
HOW TO READ IT & WHAT TO LOOK FOR
Look at the axes so you know what the major categorical ‘lanes’ represent and what the range of date values is (min to max). Then try to determine what each categorical activity line represents. As there are so many derivatives there is no single reading strategy, but generally glance across the entire chart noting the sequence of the activities; there is usually a sequential logic attached to their sorting based on the start date milestone in particular. Follow the narrative from left to right, noting observations about any big, small and medium weighted lines and spotting any moment when they connect with, overlap or detach from other activities. Are there any major convergences or divergences in pattern? Any hubs of dense activity and other sparse moments? Look for the length of lines to determine the long, medium and short durations of activity. Where available, compare the activities against annotated references about other key milestone dates that might hold some significance or influence.
PRESENTATION TIPS
ANNOTATION: Chart apparatus devices like tick marks and gridlines in particular can be helpful to increase the accuracy of the reading of both the quantitative values and the activity ‘lanes’, which may be coloured to help recognise divisions between categories. Direct
337
labelling is usually seen in these timelines to help maintain associations across the display with the categories of characters or activities, perhaps annotating the consequence or cause of lines merging, etc. Think carefully about what is the most useful and meaningful interval for your time axis labelling.
COLOUR: Even if colour does not have a direct association with given activities, it can be a useful property to highlight certain features of the narrative, sometimes acting as a container device to group activities together, even if just for a momentary time period.
COMPOSITION: Where possible, try to make the categorical sorting meaningful, maybe organising values in ascending/descending size order. The vertical (y) or horizontal (x) sequencing of time will depend on the amount of data to show and the space you have to work with. Also, depending on the narrative, the past > present ordering may be reversed.
VARIATIONS & ALTERNATIVES
There are similarities with the organic nature of the ‘alluvial diagram’, which shows ranking and quantitative change over time for a number of concurrent categories. When there are fewer inter-activity relationships and more discrete categories are involved, then the ‘Gantt chart’ offers an alternative way of showing this analysis.
Charts Activities
Gantt chart
ALSO KNOWN AS Range chart, floating bar chart
REPRESENTATION DESCRIPTION
A Gantt chart displays the start and finish points and durations for different categorical ‘activities’. The display is commonly used in
338
project management to illustrate the breakdown of a schedule of tasks but can be a useful device to show any data based on milestone dates and durations. The chart is structured around a time-based quantitative x-axis and a categorical y-axis. Each categorical activity is represented by lines positioned according to the start moment and then stretched out to the finish point. There may be several start/finish durations within the same activity row. Sometimes points are used to accentuate the start/finish positions and the line may be coloured to indicate a relevant categorical value (e.g. separating completed vs ongoing).
EXAMPLE Showing the events of birth, death and period serving in office for the first 44 US Presidents.
Figure 6.46 A Presidential Gantt Chart
HOW TO READ IT & WHAT TO LOOK FOR
Look at the axes so you know with what major categorical values each Gantt bar is associated and what the range of the date values is (min to max). Follow the narrative, noting the sequence of the categories – there is usually a sequential sorting based on the start date milestone. Glance across the entire chart and perform global comparisons to establish the high-level ranking of biggest > smallest durations (based
339
on the length of the line) as well as early and late milestones. Identify any noticeable exceptions and/or outliers. Perform local comparisons between neighbouring bars to identify proportional differences and any connected dependencies. Estimate (or read, if labels are present) the absolute values for specific categories of interest. Where available, compare the activities against annotated references about other key milestone dates that might hold some significance or influence.
PRESENTATION TIPS
ANNOTATION: Chart apparatus devices like tick marks and gridlines (or row band-shading) in particular can be helpful to increase the accuracy of the reading of the start point and duration of activities along the timeline. If you have axis labels you may not need direct labels for the values shown with each duration bar – this will be label overload, so generally decide between one or the other. Think carefully about what is the most useful and meaningful interval for your time axis labelling.
COMPOSITION: There is no significant difference in perception between vertical or horizontal Gantt charts, though horizontal layouts are more metaphorically consistent with the concept of reading time. Additionally, these layouts tend to make it easier to accommodate and read the category labels. Where possible, try to sequence the categorical ‘activities’ in a way that makes for the most logical reading, either organised by the start/finish dates or maybe the durations (depending on which has most relevance).
VARIATIONS & ALTERNATIVES
Variations might involve the further addition of different point markers (represented by combinations of symbols and/or colours) along each activity row to indicate additional milestone details, using the ‘instance chart’. An emerging trend in technique terms involves preserving the position of activity lines adjacent to other concurrent activities, rather than fixing them to stay within discrete rows. Sometimes there is much more fluidity and less ‘discreteness’ in the relationships between activity, so approaches like the ‘connected timeline’ may be more fitting.
Charts Activities
340
Instance chart
ALSO KNOWN AS Milestone map, barcode chart, strip plot
REPRESENTATION DESCRIPTION
An instance chart displays individual moments or instances of categorical ‘activities’. There are many variations in approach for this kind of display but generally you will find a structure based on a time- based quantitative x-axis and a categorical y-axis. For each categorical activity, instances of note are represented by different point markers that indicate along the timeline when something has happened. The point markers may have different combinations of symbols and colours to represent different types of occurrences, but avoid having too many different combinations so that viewers do not have to learn an entirely new alphabet of meaning.
EXAMPLE Showing the instances of different Avengers characters appearing in Marvel’s comic book titles between 1963 and 2015.
Figure 6.47 How the ‘Avengers’ Line-up Has Changed Over the Years
341
HOW TO READ IT & WHAT TO LOOK FOR
Look at the axes so you know with what major categorical values each row of instances is associated and what the range of the date values is (min to max). Look up any legend that will explain what (if any) associations exist between the instance markers and their colour/symbol. Glance down the y-axis noting the sequence of the categories; there is usually a sequential logic attached to their sorting based on the start date milestone in particular. Follow the narrative, noting observations about the type and frequency of instances being plotted. Look across the entire chart to locate the headline patterns of clustering and identify any noticeable exceptions and/or outliers. Look across the patterns within each row individually to learn about each category’s dispersal of instances. Look for empty regions where no marks appear. How do all these patterns relate to the time frame displayed? Where available, compare the activities against annotated references about other key milestone dates that might hold some significance or influence.
PRESENTATION TIPS
ANNOTATION: The main annotation properties will be used to serve the role of explaining the associations between marks and attributes
342
through clear legends/keys.
COMPOSITION: Where possible, try to sequence the categorical ‘activities’ in a way that makes for the most logical reading, either organised by the start/finish dates or maybe the durations (depending on which has most relevance).
VARIATIONS & ALTERNATIVES
Some variations may see the size of a geometric shape used instead of just a point to indicate also a quantitative measure to go with the instance. The marking of an instance through a ‘when’ moment could also be based on data that talks about positional moments within a sequence. If the basic activity is reduced to a start/finish moment then the ‘Gantt chart’ will be the best-fit option.
Charts Overlays
Choropleth map
ALSO KNOWN AS Heat map
REPRESENTATION DESCRIPTION
A choropleth map displays quantitative values for distinct, definable spatial regions on a map. Each geographic region is represented by a polygonal area based on its outline shape, with each distinct shape then collectively arranged to form the entire landscape. (Note that most tools for mapping have a predetermined reference between a region name and the dimensions of the regional polygon.) Each area is colour-coded to represent a quantitative value based on a scale with colour variation intervals that (typically) go from a light tint for smaller values to a dark shade for larger values. Choropleth maps should only be used when the quantitative measure is directly associated with and continuously relevant across the spatial region on which it will be displayed.
343
Similarly, if your quantitative measure is about or related to the consequence of more people living in an area, interpretations may be distorting, so consider transforming your data to per capita or per acre (or other spatial denominator) to standardise the analysis accordingly.
EXAMPLE Mapping the percentage change in the populations of Berlin’s districts across new and native Berliners since the fall of the Berlin Wall.
Figure 6.48 Native and New Berliners — How the S-Bahn Ring Divides the City
HOW TO READ IT & WHAT TO LOOK FOR
Acquaint yourself with the geographic region you are presented with and carefully consider the quantitative measure that is being represented. Establish the colour-scale value associations, usually found via a legend. Glance across the entire chart to locate the dark, light and medium shades (generally darker = larger) and perform global comparisons to establish the high-level ranking of biggest values > smallest. Identify any noticeable exceptions and/or outliers. Beware making judgements about the significance of prominent large
344
geographical areas: size is an attribute of the underlying region, not the significance of the measure displayed. Gradually zoom in your focus to perform increasingly local comparisons between neighbouring regional areas to identify any noticeable consistencies or inconsistencies between their values. Estimate (or read, if labels are present) the absolute values of specific regions of interest.
PRESENTATION TIPS
ANNOTATION: Directly labelling the regional areas with geographical details and the value they hold is likely to lead to too much clutter. You might include only a limited number of regional labels to provide spatial context and orientation.
COLOUR: Legends explaining the colour scales should ideally be placed as close to the map display as possible. The border colour and stroke width for each spatial area should be distinguishable to define the shape but not so prominent as to dominate attention – usually a subtle grey- or white-coloured thin stroke will be fine. As well as variation in colour scales, sometimes pattern or textures may add an extra layer of detail to the value status of each region. When including a projected mapping layer image in the background, ensure it is not overly competing for visual prominence by making it light in colour and possibly semi-transparent. Do not include any unnecessary geographical details that add no value to the spatial orientation or interpretation and clutter the display (e.g. roads, building structures).
COMPOSITION: With Earth being a sphere, there are many different mapping projections for representing the regions of the world on a plane surface. Be aware that the transformation adjustments made by some map projections can distort the size of regions of the world, inflating their size relative to other regions.
VARIATIONS & ALTERNATIVES
Some choropleth maps may be used to indicate categorical association rather than quantitative measurements. Alternative thematic mapping approaches to representing quantitative values might include the ‘proportional symbol map’ and the ‘dot density map’. This is a variation that involves plotting a representative quantity of dots equally (but randomly) across and within a defined spatial region. The position of individual dots is therefore not to be read as indicative of precise locations but used to form a measure of quantitative density. This offers a useful alternative to the choropleth map, especially when categorical separation of the dots through colour is of value. ‘Dasymetric mapping’
345
is similar in approach to choropleth mapping but breaks the constituent regional areas into much more specific, almost custom-drawn, sub- regions to better represent the realities of the distribution of human and physical phenomena within a given spatial boundary.
Charts Overlays
Isarithmic map
ALSO KNOWN AS Contour map, isopleth map, isochrone map
REPRESENTATION DESCRIPTION
An isarithmic map displays distinct spatial surfaces on a map that share the same quantitative classification. All spatial regions (transcending geo-political boundaries) that share a certain quantitative value or interval are formed by interpolated ‘isolines’ connecting points of similar measurement to form distinct surface areas. Each area is then colour-coded to represent the relevant quantitative value. The scale of colour variation intervals differs between deployments but will typically range from a light tint for smaller values to a dark shade for larger values. An isarithmic map would be used in preference to a choropleth map when the patterns of data being displayed transcend the distinct regional polygons. They could be used to show temperature bandings or smoothed regions of political attitudes.
EXAMPLE Mapping the degree of dialect similarity across the USA.
Figure 6.49 How Y’all, Youse and You Guys Talk
346
HOW TO READ IT & WHAT TO LOOK FOR
Acquaint yourself with the geographic region you are presented with and carefully consider the quantitative measure that is being represented. Establish the colour scale value associations, usually found via a legend. Glance across the entire chart to locate the dark, light and medium shades (generally darker = larger) and perform global comparisons to establish the high-level ranking of biggest values > smallest. Identify any noticeable exceptions and/or outliers, including regions that appear in isolation from their otherwise related values and notable for their position adjacent to very different shaded regions. Note that any interpolation used to smooth the joins between data points to form organic surfaces will inevitably reduce the precision of the surfaces in their relationship to land position. Gradually zoom in your focus to perform increasingly local comparisons between neighbouring regional areas to identify any noticeable consistencies or inconsistencies between their values. Estimate the absolute values of specific regions of interest.
PRESENTATION TIPS
ANNOTATION: Directly labelling the surface areas to show the quantitative value or range they represent will be too cluttered. You might include only a limited number of regional labels to provide
347
spatial context and orientation.
COLOUR: Legends explaining the colour scales should ideally be placed as close to the map display as possible. If using visible contour or boundary lines there is a clear implication of a location being inside or outside the line, so make these lines as prominent in colour as possible according to the precision of their representation. If the smoothing of the surface locations has been applied the representation of these areas should similarly avoid looking definitive. You therefore might consider subtle colour gradation/overlapping between different regions to capture appropriately the underlying ‘fuzziness’ of the data. As well as colour scales, sometimes pattern or textures may add an extra layer of detail to the value status of each surface region. When including a projected mapping layer image in the background, ensure it is not overly competing for visual prominence by making it light in colour and possibly semi-transparent. Do not include any unnecessary geographical details that add no value to the spatial orientation or interpretation and clutter the display (e.g. roads, building structures).
COMPOSITION: Be aware that the transformation adjustments made by some map projections can distort the size of regions of the world, inflating their size relative to other regions.
VARIATIONS & ALTERNATIVES
There are specific applications of isarithmic maps used for showing elevation (‘contour maps’), atmospheric pressure (‘isopleth maps’) or travel–time distances (‘isochrone maps’). Sometimes you might use isarithmic maps to show a categorical status (perhaps even a binary state) rather than a quantitative scale.
Charts Overlays
Proportional symbol map
348
ALSO KNOWN AS Graduated symbol map
REPRESENTATION DESCRIPTION
A proportional symbol map displays quantitative values for locations on a map. The values are represented via proportionally sized areas (usually circles), which are positioned with the centre mid-point over a given location coordinate. Colour is sometimes used to introduce further categorical distinction.
EXAMPLE Mapping the origin and size of funds raised across the 22 major candidates running for US President during the first half of 2015.
Figure 6.50 Here’s Exactly Where the Candidates’ Cash Came From
HOW TO READ IT & WHAT TO LOOK FOR
Acquaint yourself with the geographic region you are presented with and carefully consider the quantitative measure that is being represented. Establish the area size value associations, usually found via a legend. Glance across the entire chart to locate the large, medium and small shapes and perform global comparisons to establish the high-level ranking of biggest values > smallest. Identify any noticeable exceptions and/or outliers. Gradually zoom in your focus to perform increasingly local comparisons between neighbouring regional areas to identify any noticeable consistencies or inconsistencies between their values. Estimate (or read, if labels are present) the absolute values of specific
349
regions of interest. Also note where there are no markers. If colour is being used to further break down the categories of the values shown, identify any grouped patterns that emerge.
PRESENTATION TIPS
INTERACTIVITY: Interaction may be helpful to reveal location and value labels through selection or mouseover events.
ANNOTATION: Directly labelling the shapes with geographical details and the value they hold is likely to lead to too much clutter. You might therefore include only a limited number of regional labels to provide spatial context and orientation. Legends explaining the size scales – and any colour associations – should ideally be placed as close to the map display as possible. Avoid including unnecessary geographical details that add no value to the spatial orientation or interpretation and clutter the display (e.g. roads, building structures).
COLOUR: Sometimes the circular shapes are filled, at other times they remain unfilled. If colours are being used to distinguish the different categories, ensure these are as visibly different as possible. When a circle has a large value its shape will transgress well beyond the origin of its geographical location, intruding on and overlapping with other neighbouring values. The use of outline borders and semi-transparent colours helps with the task of avoiding occlusion (visually hiding values behind others). When including a projected mapping layer image in the background, ensure it is not overly competing for visual prominence by making it light in colour and possibly semi-transparent.
VARIATIONS & ALTERNATIVES
Variations may see the typical circle replaced by squares and geographical space replaced by anatomical regions. Alternatives to the proportional symbol map include the ‘choropleth map’, which colour- codes regions, or the ‘dot map’, which uses a dot to represent an instance of something. Avoid the temptation to turn the circle symbols into pie charts; it is not a good look. If you absolutely positively have to show a part-to-whole relationship, only show two categories, as per the recommended practice for pies.
Charts Overlays
350
Prism map
ALSO KNOWN AS Isometric map, spike map, datascape
REPRESENTATION DESCRIPTION
A prism map displays quantitative values for locations on a map. The values are represented via proportionally sized lines, appearing as 3D bars, that typically cover a fixed surface area of space and are just extended in height proportionally to represent the quantitative value for that location. Being able to judge the dimensions of 3D forms in a 2D view is very difficult, so they are only ever really used to create a gist of the profile of values, enabling recognition of the main peaks in particular.
EXAMPLE Mapping the population of trees for each 180 square km of land across the globe.
Figure 6.51 Trillions of trees
351
HOW TO READ IT & WHAT TO LOOK FOR
Acquaint yourself with the geographic region you are presented with and carefully consider the quantitative measure that is being represented. Establish the area size value associations, usually found via a legend. Glance across the entire chart to locate the large, medium and small shapes and perform global comparisons to establish the high-level ranking of biggest values > smallest. Identify any noticeable exceptions and/or outliers. Gradually zoom in your focus to perform increasingly local comparisons between neighbouring regional areas to identify any noticeable consistencies or inconsistencies between their values. Estimate (or read, if labels are present) the absolute values of specific regions of interest. Also note where there are no bars emerging from the surface.
PRESENTATION TIPS
352
INTERACTIVITY: Ideally prism maps would be provided with interactive features that allow panning around the map region to offer different viewing angles to overcome the perceptual difficulties of judging the dimensions of 3D forms in a 2D view. Without this, smaller values will be hidden behind the larger forms, just as smaller buildings are hidden by skyscrapers in a city.
ANNOTATION: Directly labelling the prism shapes is infeasible – at most you might include only a limited number of labels to provide spatial context and orientation against the largest forms. Legends explaining the size scales should ideally be placed as close to the map display as possible.
COLOUR: Most tools that enable this type of mapping will likely have visual property settings for a faux light effect, helping the physical shapes to emerge more prominently through light and shadow. Ensure colour assist in helping the shape of the forms to be as visible as possible, maybe with opacity to enable smaller values to be not entirely hidden behind any larger ones. When including a mapping layer image on the surface, ensure it is not overly competing for visual prominence by making it light in colour and possibly semi-transparent. Do not include any unnecessary geographical details that add no value to the spatial orientation or interpretation and clutter the display (e.g. roads, building structures).
COMPOSITION: Be aware that the transformation adjustments made by some map projections can distort the size of regions of the world, inflating their size relative to other regions.
VARIATIONS & ALTERNATIVES
Alternatives to the prism map, especially to avoid 3D form, include the ‘proportional symbol map’, which uses proportionally sized geometric shapes, and the ‘choropleth map’, which colour-codes regional shapes.
Charts Overlays
353
Dot map
ALSO KNOWN AS Dot distribution map, pointillist map, location map, dot density map
REPRESENTATION DESCRIPTION
A dot map displays the geographic density and distribution of phenomena on a map. It uses a point marker to indicate a categorical ‘observation’ at a geographical coordinate, which might be plotting instances of people, notable sites or incidences. The point marker is usually a filled, small dot. Colour can be used to distinguish categorical classifications. Sometimes a dot represents a one-to-one phenomenon (i.e. a single record at that location) and sometimes a dot will represent one-to-many phenomena (i.e. for an aggregated statistic whereby the location represents a logical mid-point). As the proliferation of GPS recording devices increases, the accuracy and prevalence of detailed location marked incidences are leading to increased potential for this type of approach. However, think carefully about the potential sensitivity of directly plotting a phenomenon or data incidence at a given location.
EXAMPLE Mapping each resident of the USA based on the location at which they were counted during the 2010 Census across different ethnicities.
Figure 6.52 The Racial Dot Map
354
HOW TO READ IT & WHAT TO LOOK FOR
Acquaint yourself with the geographic region you are presented with and carefully consider the phenomenon that is being represented. Establish the unit of this measure (is it a one-to-one relationship or one- to-many?) by referring to a legend. If categorical colours have been deployed, establish the different classifications and associations. Scan the chart looking for the existence of noticeable clusters as well as the widely dispersed (and maybe empty) regions. Some of the most interesting observations come from individual outliers that stand out separately from others. Are there any patterns between the presence of dots and their geographical location? Are there any patterns across the points with similar categorical colour? Gradually zoom in your focus to perform increasingly local assessments between neighbouring regional areas to identify any noticeable consistencies or inconsistencies between their patterns.
PRESENTATION TIPS
INTERACTIVITY: One method for dealing with plotting high quantities of observations is to provide interactive semantic zoom features, whereby each time a user zooms in by one level of focus, the unit quantity represented by each dot decreases, from a one-to-many towards a one-to-one relationship.
355
ANNOTATION: Direct labelling is not necessary, just provide a limited number of regional labels to offer spatial context and orientation. Legends explaining the dot unit scale and any colour associations should ideally be placed as close to the map display as possible.
COLOUR: If colours are being used to distinguish the different categories, ensure these are as visibly different as possible. When including a mapping layer image in the background, ensure it is not overly competing for visual prominence by making it light in colour and possibly semi-transparent. Do not include any unnecessary geographical details that add no value to the spatial orientation or interpretation and clutter the display (e.g. roads, building structures).
COMPOSITION: Dot maps must always be displayed on a map that demonstrates an equal-area projection as the precision of the plotted locations is paramount. From a readability perspective, try to find a balance between making the size of the dots small enough to preserve their individuality but not too tiny to be indecipherable.
VARIATIONS & ALTERNATIVES
A ‘dot density map’ is a variation that involves plotting a representative quantity of dots equally (but randomly) across and within a defined spatial region. The position of individual dots is therefore not to be read as indicative of precise locations but used to form a measure of quantitative density. This offers a useful alternative to the choropleth map, especially when categorical separation of the dots through colour is of value. Plotting the location of an incidence of a phenomenon can transcend geographical mapping to any spatial display, such as the seat layout and availability at a theatre or on a flight, or showing the key patterns of play across a sports pitch.
Charts Overlays
Flow map
356
ALSO KNOWN AS Connection map, route map, stream map, particle flow map
REPRESENTATION DESCRIPTION
A flow map shows the characteristics of the movement or flow of a phenomenon across spatial regions. It is often formed using line marks to map flow and combinations of attributes to display the characteristics of this flow. Examples might include the patterns of traffic and travel across or between given routes, the dynamics of the patterns of weather, or the movement patterns of people or animals. There is no fixed template for a flow map but it generally displays characteristics of origin and destination (positions on a map), route (using organic or vector paths), direction (arrow or tapered line width), categorical classification (colour) and some quantitative measure (line weight or motion speed).
EXAMPLE Mapping the average number of vehicles using Hong Kong’s main network of roads during 2011.
Figure 6.53 Arteries of the City
357
HOW TO READ IT & WHAT TO LOOK FOR
Acquaint yourself with the geographic region you are presented with and carefully consider the phenomenon that is being displayed. Establish the association of all visible attributes to understand fully their classification and representation, such as the use of quantitative scales (colour, line size or width) or categorical associations (colour). Scan the chart looking for the existence of patterns of movement, maybe through clustering or common direction, and identify any main hubs and densities within the network. Find the large and the small, the dense and the sparse, and draw out any patterns formed by colour classifications. Gradually zoom in your focus to perform increasingly local assessments between neighbouring regional areas to identify any noticeable consistencies or inconsistencies between their patterns.
PRESENTATION TIPS
INTERACTIVITY: Animated sequences will be invaluable to convey motion if the nature of the flow being presented has the relevant physics of movement.
358
ANNOTATION: Annotation needs will be unique to each approach and the inherent complexity or otherwise of the display. Often the general patterns may offer the sufficient level of readability without the need for imposing amounts of value labels.
COLOUR: The colour relationship needs careful consideration to get the right balance between the intricacies of the foreground data layer and the background mapping layer image. Ensure the background is not overly competing for visual prominence by making it light in colour and possibly semi-transparent. Do not include any unnecessary geographical details that add no value to the spatial orientation or interpretation, but do include those features that have a direct association with the subject matter (such as roads, routes, etc.).
COMPOSITION: Some degree of geographic distortion of routes or connecting lines may be required practically to display flow data. Choices like interpolation of lines to smooth an activities route or the merging of relatively similar pathways may be entirely legitimate but ensure that this is made clear to the reader.
VARIATIONS & ALTERNATIVES
There are naturally many variations in how you might show flow. It generally differs between whether you are showing point A to point B ‘connection maps’, more nuanced ‘route maps’ or surface phenomena such as ‘particle flow maps’.
Charts Distortions
Area cartogram
ALSO KNOWN AS Contiguous cartogram, density-equalizing map
EXAMPLE Mapping the measures of climate change responsibility compared to vulnerability across all countries.
359
REPRESENTATION DESCRIPTION
An area cartogram displays the quantitative values associated with distinct definable spatial regions on a map. Each geographic region is represented by a polygonal area based on its outline shape with the collective regional shapes forming the entire landscape. (Note that most tools for mapping have a predetermined reference between a region name and the dimensions of the regional polygon.) Quantitative values are represented by proportionately distorting (inflating or deflating) the relative size of and, to some degree, shape of the respective regional areas. Traditionally, area cartograms strictly aim to preserve the neighbourhood relationships between different regions. Colour is sometimes used to further represent the same quantitative value or to associate the region with a categorical classification. Area cartograms require the reader to be relatively familiar with the original size and shape of regions in order to be able to establish the degree of relative change in their proportions. Without this it is almost impossible to assess the degree of distortion and indeed to identify the regions themselves.
Figure 6.54 The Carbon Map
HOW TO READ IT & WHAT TO LOOK FOR
Acquaint yourself with the geographic region you are presented with and carefully consider the quantitative measure that is being represented. Establish the quantitative value scales or categorical classifications associated with the colour scale, usually found via a legend. Glance across the entire chart to locate the big-, small- and medium-sized shapes according to their apparent distortion. Identify any noticeable exceptions and/or outliers. Gradually zoom in your focus
360
to perform increasingly local comparisons between neighbouring regional areas to identify any noticeable consistencies or inconsistencies between their values. Estimate (or read, if labels are present) the absolute values of specific regions of interest.
PRESENTATION TIPS
INTERACTIVITY: Animated sequences enabled through interactive controls can help to better identify instances and degrees of change but usually only over a small set of regions and only if the change is relatively smooth and sustained. Manual animation will help provide more control over the experience.
ANNOTATION: Directly labelling the regional areas with geographical details and the value they hold is likely to lead to too much clutter. You might include only a limited number of regional labels to provide spatial context and orientation.
COLOUR: Legends explaining any colour scales should ideally be placed as close to the map display as possible. The border colour and stroke width for each spatial area should be distinguishable to define the shape but not so prominent as to dominate attention, usually a subtle grey- or white-coloured thin stroke will be fine.
COMPOSITION: To aid the readability of the size of the distortions, it can be useful to present a thumbnail view of the undistorted original geographical layout to help the readers orient themselves with the changes.
VARIATIONS & ALTERNATIVES
Unlike contiguous cartograms, non-contiguous cartograms tend to preserve the shape of the individual polygons but modify the size and the neighbouring connectivity to other adjacent regional polygon areas. The best alternative ways of showing similar data would be to consider using the ‘choropleth map’ or ‘Dorling cartogram’.
Charts Distortions
361
Dorling cartogram
ALSO KNOWN AS Demers cartogram
REPRESENTATION DESCRIPTION
A Dorling cartogram displays the quantitative values associated with distinct, definable spatial regions on a map. Each geographic region is represented by a circle which is proportionally sized to represent a quantitative value. The placement of each circle generally resembles the region’s geographic location with general preservation of neighbourhood relationships between adjacent shapes. Colour is used to associate the region with a categorical classification.
EXAMPLE Mapping the predicted electoral voting results for each state in the 2012 Presidential Election.
Figure 6.55 Election Dashboard
362
HOW TO READ IT & WHAT TO LOOK FOR
Acquaint yourself with the geographic region you are presented with and carefully consider the quantitative measure that is being represented. Establish the quantitative value scales or categorical classifications associated with the colour scale, usually found via a legend. Glance across the entire chart to locate the big-, small- and medium-sized shapes. Identify any noticeable exceptions and/or outliers. Gradually zoom in your focus to perform increasingly local comparisons between neighbouring regional areas to identify any noticeable consistencies or inconsistencies between their values. Estimate (or read, if labels are present) the absolute values of specific regions of interest.
PRESENTATION TIPS
INTERACTIVITY: Interactive features that enable annotation for
363
category and value labelling can be useful to overcome the difficulties associated with the geographic distortion.
ANNOTATION: Directly labelling the shapes with geographical details and the value they hold is common, though you might restrict this to the circles that have sufficient size to hold such annotation. Otherwise you will need to decide how to handle the labelling of small values.
COLOUR: Legends explaining the size scales and colour associations should ideally be placed as close to the map display as possible. If colours are being used to distinguish the different categories, ensure these are as visibly different as possible.
COMPOSITION: Remember that preserving the adjacency with neighbouring regions is important. Dorling cartograms tend not to allow circles to overlap or occlude, so some accommodation of large values might result in location distortion.
VARIATIONS & ALTERNATIVES
A variation on the approach, called the ‘Demers cartogram’, involves the use of squares or rectangles instead of circles, which offers an alternative way of connecting adjacent shapes. Other approaches would be through the ‘area cartogram’ and the ‘choropleth map’.
Charts Distortions
Grid map
ALSO KNOWN AS Cartogram, bin map, equal-area cartogram, hexagon bin map
REPRESENTATION DESCRIPTION
A grid map displays the quantitative values associated with distinct
364
definable spatial regions on a map. Each geographic region (or a statistically consistent interval of space, known as a ‘bin’) is represented by a fixed-size uniform shape, sometimes termed a ‘tile’. The shapes used tend to be squares or hexagons, though any tessellating shape would work in theory in order to help arrange all the regional tiles into a collective shape that roughly fits the real-world geographical adjacency. Colours are applied to each regional tile either to represent a quantitative value or to associate the region with a categorical classification. Note that the mark used for this chart type is a point rather than an area mark as its size attributes are constant.
EXAMPLE Showing the percentage of household waste recycled in each council region across London between April 2013 to March 2014.
Figure 6.56 London is Rubbish at Recycling and Many Boroughs are Getting Worse
HOW TO READ IT & WHAT TO LOOK FOR
365
Acquaint yourself with the geographic region you are presented with and carefully consider the quantitative measure that is being represented. Identify the general layout of the constituent tiles to determine how good a fit they are with their adjacent regions in absolute and relative geographical terms. Establish the categorical or quantitative classifications associated with the colour scale, usually found via a legend. Glance across the entire chart to locate the big, small and medium shaded tiles (if quantitative) or the main patterns formed by the categorical colouring. Identify any noticeable exceptions and/or outliers. Gradually zoom in your focus to perform increasingly local comparisons between neighbouring regional areas to identify any noticeable consistencies or inconsistencies between their values. Estimate (or read, if labels are present) the absolute values of specific regions of interest.
PRESENTATION TIPS
INTERACTIVITY: Interactive features that enable annotation for category and value labelling can be useful to overcome the difficulties associated with the geographic distortion.
ANNOTATION: Directly labelling the shapes with geographical details is usually too hard. Some versions of the ‘grid map’ will include abbreviated labels, maybe two digits, to indicate the region they represent and to aid orientation. Otherwise it may require interactivity to facilitate such annotations. Legends explaining the colour associations should ideally be placed as close to the map display as possible.
COLOUR: If colour is being used to distinguish the different categories, ensure they are as visibly different as possible.
COMPOSITION: The main challenge is to find the most appropriate and representative tile–region relationship (what is the right amount and geographical level for each constituent tile?) and to optimise the best-fit collective layout that preserves as many of the legitimate neighbouring regions as possible.
VARIATIONS & ALTERNATIVES
‘Hexagon bin maps’ are specific deployments of the grid map that offer a layout formed by a high resolution of smaller hexagons to preserve localised details. Beyond geographical space, the grid map approach is applicable to any spatial analysis such as in sports.
366
6.3 Influencing Factors and Considerations Having covered the fundamentals of visual encoding and profiled many chart type options that deploy different encoding combinations you now need to consider the general factors that will influence your specific choices for which chart or charts to use for your data representation.
Choosing which chart type(s) to use is, inevitably, not a single-factor decision. Rather, as ever with data visualisation, it is an imperfect recipe made up of many ingredients. A pragmatic balance has to be found somewhere between taking on board the range of influencing factors that shape selections and not becoming frozen with indecision caused by the burden of having to consider so many different issues.
Firstly, you need to reflect on the relevant factors that emerge from the first three ‘preparatory’ stages of the design process and then supplement this by addressing the guidance offered by the three visualisation design principles introduced in Chapter 1. It must be emphasised that there are no direct answers provided for you here, simply guidance. How you might resolve the unique challenges posed by your project has to be something you arrive at yourself.
Formulating Your Brief Skills and resources, frequency: What charts can you actually make and how efficiently can you create them? This is the big question. Having the ability to create a broad repertoire of different chart types is the vocabulary of this discipline, judging when to use them is the literacy. What will have a great influence on the ambitions of the type of charts you might employ is the ‘expressiveness’ of your abilities and that of the technology (applications, programs, tools) you have access to. Expressiveness is a term I first heard used in this context by Arvind Satyanarayan, a Computer Science PhD candidate at Stanford University. It describes the amount of variety and extent of control you are provided with by a given technology in the construction of your visualisation solution, so long as you also possess the necessary skills to exploit such features, of course:
In a data representation context, maximum expressiveness means you can create any combination of mark and attribute encoding
367
to display your data – that is, you can create many different charts. Programming libraries like D3.js and open source tools like R offer broad libraries of different chart options and customisations. The drawing-by-hand nature of Adobe Illustrator would similarly enable you to create a wide range of solutions (though unquestionably more manual in effort and less replicable). Restricted expressiveness means you have much more limited scope to adapt different mark and attribute encodings. Indeed you might be faced with assigning data to the fixed encoding options afforded by a modest menu of chart types. A tool like Excel has a relatively limited range of (useful) chart types in its menu. While there are ways of enhancing the options through plugins and different ‘workaround’ techniques that broaden its scope, it is a relatively limited tool. It may, however, suffice for most people’s visualisation ambitions. Elsewhere, there are many web-based visualisation creation tools which are of value for those who want quick and simple charting, though they certainly reduce the range of options and the capability to customise their appearance.
‘The capability to cope with the technological dimension is a key attribute of successful students: coding - more as a logic and a mindset than a technical task - is becoming a very important asset for designers who want to work in Data Visualization. It doesn’t necessarily mean that you need to be able to code to find a job, but it helps a lot in the design process. The profile in the (near) future will be a hybrid one, mixing competences, skills and approaches currently separated into disciplinary silos.’ Paolo Ciuccarelli, discussing students on his Communication Design Master Programme at Politecnico di Milano
As you reflect on the gallery of charts, my advice would be to perform an assessment of the charts you can make using a scoring system as follows:
368
For any of the charts that fail to score 3 points, here are some strategies to dealing with this:
Tools are continually being enhanced. The applications you use now that cannot create, for example, a Sankey diagram, may well offer that in the next release. So wait it out! For those charts that currently score 1 or 0 points, look around the web for examples of workaround approaches that will help you achieve them. For example, you might use conditional formatting in an Excel worksheet to create a rudimentary heat map. This is not a chart type offered as standard within the tool but represents an innovative solution through appropriating existing features intended to serve other purposes. Any such solutions, though, have to be framed by the frequency of your work – will this work realistically need to be replicable and repeatable (for example, every month) and does my solution make that achievable? Invest time in developing skills in the other tools to broaden your repertoire. Tools like R have a large community of users sharing code, tutorials and examples, resources that would greatly help to facilitate your learning. Lower your ambitions. Sometimes the most significant discipline to demonstrate is acknowledging what you cannot do and accepting that (at least, for now) you might need to sacrifice the ideal choices you would make for more pragmatic ones.
Purpose: Should you even seek to represent you data in chart form? Will it add any value, enabling new insights or greater perceptual efficiency compared with its non-visualised form? Will portraying your data via an elegantly presented table, offering the viewer the ability to look up and reference values, actually offer a more suitable solution? Do not rule out the value of a table. Additionally, perhaps you are trying to represent something in chart form that would actually be better displayed through
369
information-based (rather than data-based) explanations using imagery, textual anecdotes, video and photos? Most of the time the charting of data will be fit for purpose, but just keep reminding yourself that you do not have to chart everything – just make sure you are doing it to add value.
‘I was in the middle of this huge project, juggling as fast and as focused as I could, and I had this idea of a set of charts stuck in my head that kept resurfacing. And then, as we were heading close to deadline, I realized I couldn’t do it. I failed. I couldn’t make it work. Because we had pictures of the children, and that was enough … I had to let it go.’ Sarah Slobin, Visual Journalist, discussing a project profiling a group of families with children who have a fatal disease
Purpose map: In defining the ‘tone’ of the project, your were determining what the optimum perceptibility of your data would be for your audience. Your definitions were based on whether you were aiming to facilitate the reading of the data or more a general feeling of the data? Were you concerned with enabling precise and accurate perceptions of values or is it more about the sense-making of the big, medium and small judgments – getting the ‘gist’ of values more than reading back the values? Were there emotional qualities that you wanted to emphasise perhaps at the compromise of perceptual efficiency? Maybe there was a balance between the two?
How these tonal definitions apply specifically to data representation requires our appreciation of some fundamental theory about data visualisation. In his book Semiology Graphique, published in 1967, Jacques Bertin was the first, most notable author to propose the idea that different ways of encoding data might offer varying degrees of effectiveness in perception. In 1984 William Cleveland and Robert McGill published a seminal paper, ‘Graphical Perception: Theory, Experimentation, and Application to the Development of Graphical Methods’, that offered more empirical evidence of Bertin’s thoughts. They produced a general ranking that explained which attributes used to encode quantitative values would facilitate the highest degree of perceptual accuracy. In 1986, Jock Mackinlay’s paper, ‘Automating the Design of Graphical Presentations of Relational Information’, further extended this to include proposed rankings for encoding categorical nominal and categorical ordinal data types as well as quantitative ones. The table shown in Figure 6.57, adapted from Mackinlay’s paper, presents the ‘Ranking of
370
Perceptual Tasks’.
In a nutshell, this ancestry of studies reveals that certain attributes used to encode data may make it easier, and others may make it harder, to judge accurately the values being portrayed. Let’s illustrate this with a couple of examples. Looking at Figure 6.58, ask yourself: if A is 10, how big is B in the respective bar and circular displays?
In both cases the answer is B = 5, but while the B ‘bar’ being 5 feels about right, the idea that the B ‘circle’ is 5 does not feel quite right. That is because our ability to perform relative judgements for the length of bars is far more precise and accurate than the relative judgements for the area of circles. This is explained by the fact that when judging the variation in size of a line (bar) you are detecting change in a linear dimension, whereas the variation in size of a geometric area (circle) occurs across a quadratic dimension. If you look at the rankings in Figure 6.57 in the ‘Quantitative’ column, you will see the encoding attribute of Length is ranked higher than the attribute of Area.
Figure 6.57 The Ranking of Perceptual Tasks
371
Figure 6.58 Comparison of Judging Line Size vs Area Size
Now let’s consider an example (Figure 6.59) that shows the relative accuracy of using different dimensions of colour variation to represent categorical nominal values. In the next pair of charts you can see different attributes being used to represent the categorical groupings of the points in the respective scatter plots. On the left you can see variation in the attribute of colour hue (blue, orange and green) to separate the categories visually; on the right you will see the attribute of shape (diamond, circle and square) applied to the same category groupings. What you should be experiencing is a far more immediate, effortless and accurate sense of the groupings of the coloured category markers compared with the shaped
372
category markers. It is simply easier to spot the associations through variation in colour than variation in shape. This explains why colour hue is much higher in the proposed rankings for nominal data than shape.
Figure 6.59 Comparison of judging related items using variation in colour (hue) vs variation in shape
So you can see from these simple demonstrations that there are clearly ways of encoding data that will make it easier to read values accurately and efficiently. However, as Cleveland and McGill stress in their paper, this ranking should be taken as only one ingredient of guidance: ‘The ordering does not result in a precise prescription for displaying data but rather is a framework within which to work’.
This is important to note because you have to take into account other factors. You have to decide whether precise perceiving is actually what you need to facilitate for your readers. If you do, then the likes of the bar chart – through the variation in length of a bar – will evidently offer a very precise approach. As stated in Chapter 3, that is why they are such an important part of your visual artillery.
However, sometimes getting a ‘gist’ of the data is sufficient. A few pages ago I presented an image of a bubble chart on my website’s home page, showing the popularity of my blog posts over the previous 100-day period. The purpose of this display was purely to give visitors a sense of the general order of magnitude from the most popular to the relative least popular posts. I do not need visitors to form a precise understanding of absolute values or exact rankings. I just want them to get a sense of the ranking hierarchy. I can therefore justify moving down the quantitative attribute rankings proposed and deploy a series of circles that encode the
373
visitor totals through the size of their area (colour is used to represent different article categories). The level of perceptibility (accuracy and efficiency) that I need to facilitate is adequately achieved by the resulting ‘frogspawn’-like display. Furthermore, it offers an appealing and varied display that suits the purpose of this front-page navigation device.
In practice, what all this shows is that chart types vary in the relative efficiency and accuracy of perception offered to a viewer. Moreover, many of the charts shown in the gallery can therefore only ever facilitate a gist of the values of data due to the complexity of their mark and attribute combinations and the amount of data values they might typically contain (e.g. the treemap often has many parts of a whole in a single display). It is up to you to judge what the right threshold is for your purpose.
Working With Data Data examination: Inevitably, the physical characteristics of your data are especially influential. What types of data you are trying to display will have a significant impact on how you are able to show them. Only certain types of data can fit into certain chart types; only certain chart types can accommodate certain types of data. That is why it is often most useful practically to think of this task in terms of chart types and particularly in terms of these as templates, able to accommodate specific types of data. For example, representing data through a bar chart requires one categorical variable (e.g. department) and one quantitative variable (e.g. maximum age). If you want to show a further categorical variable (let’s say, to break down departments by gender) you are going to need to switch ‘template’ and use something like a clustered bar chart which can accommodate this extra dimension. I explained earlier how the shape of data influenced the viability of the flower metaphor used in the ‘Better Life Index’. The range of categorical and quantitative values will certainly influence the most appropriate chart type choice. For example, suppose you want to show some part-to-whole analysis and you have only three parts (three sub-categories belonging to the major category or whole) then a treemap really does not make a great deal of sense – they are better at representing many parts to a whole. The unloved pie chart would probably suffice if the percentage values were quite diverse otherwise the bar chart would be best.
374
Beyond the size and shape of your data you also might be influenced by its inherent meaning. Sometimes, you will have scope in your encoding choices to incorporate a certain amount of visual immediacy in accordance with your topic. The flowers of the Better Life Index feel consistent in metaphor with the idea of better life: the more in bloom the flowers, the more colourful and proud each petal appears and the better the quality of life in that country. There is a congruence between subject matter and visual form. Think about the billionaires’ project from earlier in the chapter, with rankings displayed by industry. Each point marking each billionaire was a small caricature face. This is not necessary – a small circular mark for each person would have been fine – but by using a face for the mark it creates a more immediate recognition that the subject matter is about people. Data exploration: One consistently useful pointer towards how you might visually communicate your data to others is to consider which techniques helped you to unearth key insights when you were visually exploring the data. What chart types have you already tried out and maybe found to reveal interesting patterns? Exploratory data analysis is, in many ways, a bridge to visual communication: the charts you use to inform yourself often represent prototype thinking on how you might communicate with others. The design execution may end up being different once you introduce the influence of audience characteristics into your thinking, naturally, but if a method is already working, why not utilise the same approach again?
‘Effective graphics conform to the Congruence Principle according to which the content and format of the graphic should correspond to the content and format of the concepts to be conveyed.’ Barbara Tversky and Julie Bauer Morrison, taken from Animation: Can it Facilitate?
Establishing Your Editorial Thinking Angle: When articulating the angles of analysis you intend to portray to your viewers, you are effectively dictating which chart types might be most relevant. If you intend to show how quantities have changed over time, for example, there will be certain charts best placed to portray that and many others that will not. By expressing your desired editorial angles of analysis in language terms, this will be extremely helpful in identifying the primary families of charts across the
375
CHRTS taxonomy that will provide the best option. It is vital to treat every representation challenge on its own merits – do not fall into the trap of going through the motions. Just because you have spatial data does not mean that the most useful portrayal of that data will be via a map. If the interesting insights are not regionally and spatially significant, then the map may not provide the most relevant window on that data. The composition of a map – the shape, size and positioning of the world’s regions – is so diverse, inconsistent and truly non-uniform that it may hinder your analysis rather than illuminate it. So always make sure you have carefully considered the relevance of your chosen angle through your editorial thinking.
Trustworthy Design Avoiding deception: In the discussion about tone I explained how variations in the potential precision of perception may be appropriate for the purpose and context of your work. Precision in perception is one thing, but precision in design is another. Being truthful and avoiding deception in how you portray data visually are fundamental obligations. There are many ways in which viewers can be deceived through incorrect and inappropriate encoding choices. The main issues around deception tend to concern encoding the size of quantities. For beginners, these mistakes can be entirely innocent and unintended but need to be eradicated immediately.
Geometric calculations – When using the area of shapes to represent different quantitative values, the underlying geometry needs to be calculated accurately. One of the common mistakes when using circles, for example, is simply to modify the diameters: if a quantitative value increases from 10 to 20, just double the diameter, right? Wrong. That geometric approach would be a mistake because, as viewers, when perceiving the size of a circle, it is the area, not the width, of the circle upon which we base our estimates of the quantitative value being represented.
Figure 6.60 Illustrating the correct and incorrect circle size encoding
376
The illustration in Figure 6.60 shows the incorrect and correct ways of encoding two quantitative values through circle size, where the value of A is twice the size of B. The orange circle for B has half the diameter of A, the green circle for B has half the area of A. The green circle area calculations are the correct way to encode these two values, whereas the orange circle calculations disproportionately shrink circle B by halving the diameter rather than halving the area. This makes it appear much smaller than its true value.
3D decoration – In the vast majority of circumstances the use of 3D charts is at best unnecessary and at worst hugely distorting in the display of data. I have some empathy for those who might volunteer that they have made and/or like the look of 3D charts. In the past I did too. Sometimes we don’t know not to do something until we are told. So this is me, here and now, telling you.
The presence of 3D in visualisation tends to be motivated by a desire to demonstrate technical competence with the features of a tool in terms of ‘look how many things I know how to do with this tool!’ (users of Excel, I am pointing an accusatory finger at you right now). It is also driven by the appetite of rather unsophisticated viewers who are still attracted by the apparent novelty of 3D skeuomorphic form. (Middle and senior management of the corporate world, with your ‘make me a fancy chart’ commands, my finger of doom is now pointing in your direction.)
Using psuedo-3D effects in your charts when you have only two dimensions of data means you are simply decorating data. And when I say ‘decorating’, I mean this with the same sneer that
377
would greet memories of avocado green bathrooms in 1970s Britain. A 3D visualisation of 2D data is gratuitous and distorts the viewer’s ability to read values within any degree of acceptable accuracy. As illustrated in Figure 6.61, in perceiving the value estimates of the angles and segments in the respective pie charts, the 3D version makes it much harder to form accurate judgements. The tilting of the isometric plane amplifies the front part of the chart and diminishes the back. It also introduces a raised ‘step’ which is purely decorative, thus embellishing the judgement of the segment sizes.
Figure 6.61 Illustrating the Distortions Created by 3D Decoration
Furthermore, for charts based on three dimensions of data, 3D effects should only be considered if – and only if – the viewer is provided with means to move around the chart object to establish different 2D viewing angles and the collective representation of all the 3D of data makes sense in showing a whole ‘system’. Truncated axis scales – When quantitative values are encoded through the height or length components of size (e.g. for bar charts and area charts), truncating the value axis (not starting the range of quantitative values from the true origin of zero) distorts the size judgements. I will look at this in more detail in the chapter on composition because it is ultimately more about the size considerations of scales and deployment of chart apparatus than necessarily just the representation choices.
Accessible Design
378
The bullet chart is a derivative of the bar chart – the older, more sophisticated brother of the idiot gauge chart – but I didn’t think it was necessary to profile as a separate chart type.
Encoded overlays: Beyond the immediate combinations of marks and attributes that comprise a given chart type, you may find value in incorporating additional detail to help viewers with the perceiving and interpretation task. Encoded overlays are useful to help explain further the context of values and amplify the interpretation of the good and the bad, the normal and the exceptional. In some ways these features might be considered forms of annotation, but as they represent data values (and therefore require encoding choices) it makes sense to locate these options within this chapter. There are many different types of visual overlays that may be useful to include:
Figure 6.62 Example of a Bullet Chart Using Banding Overlays
Figure 6.63 Excerpt from ‘What’s Really Warming the World?’
Bandings – These are typically shaded areas that provide some
379
sense of contrast between the main data value marks and contextual judgements of historic or expected values. In a bullet chart (Figure 6.62) there are various shaded bands that might help to indicate whether the bar’s value should be considered bad, average or good. In the line chart (Figure 6.63) here you can see the observed rise in global temperatures. To facilitate comparison with potentially influencing factors, in the background there is a contextual overlay showing the change in greenhouse gases with banding to indicate the 95% confidence interval. Markers – Adding points to a display might be useful to show comparison against a target, forecast, a previous value, or to highlight actual vs budget. Figure 6.64 shows a chart that facilitates comparisons against a maximum value marker.
Figure 6.64 Example of Using Markers Overlays
Figure 6.65 Why Is Her Paycheck Smaller?
380
Reference lines – These are useful in any display that uses position or size along an axis as an attribute for a quantitative value. Line charts or scatter plots (Figure 6.65) are particularly enhanced by the inclusion of reference lines, helping to direct the eye towards calculated trends, constants or averages and, with scatter plots specifically, the lines of best fit or correlation.
Elegant Design Visual appeal: This fits again with the thinking about ‘tone’ and may also be informed by some of the mental visualisations that might have formed in the initial stages of the process. Although you should not allow yourself to be consumed by ideas over the influence of the data, sometimes there is scope to squeeze out an extra sense of stylistic association between the visual and the content. For example, the ‘pizza’ pie chart in Figure 6.66 presents analysis about the political contributions made by companies in the pizza industry. The decision to use pizza slices as the basis of a pie chart makes a lot of sense. The graphic in Figure 6.67 displays the growth in online sales of razors. Like the pizzas, the notion of creating bar charts by scraping away lengths of shaving foam offers a clever, congruent and charming solution.
Figure 6.66 Inside the Powerful Lobby Fighting for Your Right to Eat
381
Pizza
Figure 6.67 Excerpt from ‘Razor Sales Move Online, Away From Gillette’
382
Summary: Data Representation Visual Encoding All charts are based on combinations of marks and attributes:
Marks: represent records (or aggregation of records) and can be points, lines, areas or forms. Attributes: represent variable values held for each record and can include visual properties like position, size, colour, connection.
Chart Types If visual encoding is the fundamental theoretical understanding of data representation, chart types are the practical application. There are five families of chart types (CHRTS mnemonic):
383
Influencing Factors and Considerations
Formulating the brief: skills and resources – what charts can you make and how efficiently? From the definitions across the ‘purpose map’ what ‘tone’ did you determine this project might demonstrate? Working with data: what is the shape of the data and how might that impact on your chart design? Have you already used a chart type to explore your data that might prove to be the best way to communicate it to others? Establishing your editorial thinking: what is the specific angle of the enquiry that you want to portray visually? Is it relevant and representative of the most interesting analysis of your data? Trustworthy design: avoid deception through mistaken geometric calculations, 3D decoration, truncated axis scales, corrupt charts. Accessible design: the use of encoded overlays, such as bandings, markers, reference lines, can aid readability and interpretation. Elegant design: consider the scope of certain design flourishes that might enhance the visual appeal through the form of your charts whilst also preserving their function.
Tips and Tactics
Data is your raw material, not your ideas, so do not arrive at this stage desperate and precious about wanting to use a certain data representation approach. Be led by the preparatory work (stages 1 to 3) but do use the chart type gallery for inspiration if you need to unblock! Be especially careful in how you think about representing instances of zero, null (no available data) and nothing (no observation). Do not be too proud to acknowledge when you have made a bad call or gone down a dead end.
384
385
7 Interactivity
The advancement of technology has entirely altered the nature of how we consume information. Whereas only a generation ago most visualisations would have been created exclusively for printed consumption, developments in device capability, Internet access and bandwidth performance have created an incredibly rich environment for digital visualisation to become the dominant output. The potential now exists for creative and capable developers to produce powerful interactive and engaging multimedia experiences for cross-platform consumption.
Unquestionably there is still an fundamental role for static (i.e. not interactive) and print-only work: the scope offered by digital simply enables you to extend your reach and broaden the possibilities. In the right circumstances, incorporating features of interactivity into your visualisation work offers many advantages:
It expands the physical limits of what you can show in a given space. It increases the quantity and broadens the variety of angles of analysis to serve different curiosities. It facilitates manipulations of the data displayed to handle varied interrogations. It increases the overall control and potential customisation of the experience. It amplifies your creative licence and the scope for exploring different techniques for engaging users.
The careful judgements that distinguish this visualisation design process must be especially discerning when handling this layer of the anatomy. Well-considered interactivity supports, in particular, the principle of ‘accessible’ design, ensuring that you are adding value to the experience, not obstructing the facilitation of understanding. Your main concern in considering potential interactivity is to ensure the features you deploy are useful. This is an easy thing to say about any context but just because you can does not mean to say you should. For some who possess a natural technical flair, there is often too great a temptation to create interactivity where it is neither required nor helpful.
386
Having said that, beyond the functional aspects of interactive design thinking, depending on the nature of the project there can be value attached to the sheer pleasure created by thoughtfully conceived interactive features. Even if these contribute only ornamental benefit there can be merit in creating a sense of fun and playability so long as such features do not obstruct access to understanding.
There is a lot on your menu when it comes to considering potential interaction design features. As before, ahead of your decision making about what you should do, you will first consider what you could do. To help organise your thinking, your options are divided into two main groups of features:
Data adjustments: Affecting what data is displayed. Presentation adjustments: Affecting how the data is displayed.
There is an ever-increasing range of interfaces to enable interaction events beyond the mouse/touch through gesture interfaces like the Kinect device, oculus rift, wands, control pads. These are beyond the scope of this book but it is worth watching out for developments in the future, especially with respect to the growing interest in exploring the immersive potential of virtual reality (VR).
When considering potential interactive features you first need to recognise the difference between an event, the control and the function. The event is the input interaction (such as a click), applied to a control (maybe a button) or element on your display, with the function being the resulting operation that is performed (filter the data).
Where once we were limited to the mouse or the trackpad as the common peripheral, over the past few years the emergence of touch-screens in the shape of smartphones and tablets has introduced a whole new event vocabulary. For the purposes of this chapter we focus on the language of the mouse or trackpad, but here is a quick translation of the equivalent touch events. Note that arguably the biggest difference in assigning events to interactive data visualisations exists in the inability to register a mouseover (or ‘hover’) action with touch-screens.
387
7.1 Features of Interactivity: Data Adjustments This first group of interactive features covers the various ways in which you can enable your users to adjust and manipulate your data. Specifically, they influence what data is displayed at a given moment.
I will temporarily switch nomenclature to ‘user’ in this chapter because a more active role is needed than ‘viewer’.
Framing: There is only so much one can show in a single visualisation display and thus giving users the ability to modify criteria to customise what data is visible at any given point is a strong advantage. Going back to the discussion on editorial thinking, in Chapter 5, this set of adjustments would specifically concern the ‘framing’ of what data to isolate, include or exclude from view.
For those of you familiar with databases, think of this group of features as similar in scope to modifying the criteria when querying data in a database.
388
In ‘Gun Deaths’ (Figure 7.1), you can use the filters in the pop-up check-box lists at the bottom to adjust the display of selected categorical data parameters. The filtered data is then shown in isolation above the line from all non-selected groups, which are shown below the line. The ‘Remove filters’ link can be used to reset the display to the original settings.
Figure 7.1 US Gun Deaths
In the bubble map view of the ‘FinViz’ stock market analysis site, you can change the values of the handles along the axes to modify the maximum and minimum axis range, which allows you effectively to zoom in on the records that match this criterion. You can also select the dropdown menus to change the variables plotted on each axis.
Notice the subtle transparency of the filter menu (in Figure 7.1) so that it doesn’t entirely occlude the data displayed beneath.
389
Figure 7.2 FinViz: Standard and Poor’s 500 Index
Navigating: There are dynamic features that enable users to expand or explore greater levels of detail in the displayed data. This includes lateral movement and vertical drill-down capabilities.
You will see that many of these interactive projects include links to share the project (or view of the project) with others via social media or through offering code to embed work into other websites. This helps to mobilise distribution and open up wider access to your work.
Figure 7.3 The Racial Dot Map
390
The dot map in Figure 7.3, showing the 2010 Census data, displays population density across the USA. As a user you can use a scrollable zoom or scaled zoom to zoom in and out of different map view levels. The map can also be navigated laterally to explore different regions at the same resolution. This act of zooming to increase the magnification of the view is known as a geometric zoom. This is considered a data adjustment because through zooming you are effectively re-framing the window of the included and excluded data at each level of view. In the ‘Obesity Around the World’ visualisation (Figure 7.4), selecting a continent connector expands the sub-category display to show the marks for all constituent countries. Clicking on the same connector collapses the countries to revert back to the main continent- level view. The ‘Social Progress Imperative’ project (Figure 7.5) provides an example of features that enable users to view the tabulated form of the data – the highest level of detail – by selecting the ‘Data Table’ tab. The data adjustment taking place here is through providing access to the data in a non-visual form. Users can also export the data
391
by clicking on the relevant button to conduct further local analysis.
Figure 7.4 Obesity Around the World
Animating: Data with a temporal component often lends itself to being portrayed via animated sequences. The data adjustment taking place here involves the shifting nature of the timeframe in view at any given point. Operations used to create these sequences may be automatic and/or manual in nature.
Figure 7.5 Excerpt from ‘Social Progress Index 2015’
392
This next project (Figure 7.6) plots NFL players’ height and weight over time using an animated heat map. When you land on the web page the animation automatically triggers. Once completed, you can also select the play button to recommence the animation as well as moving the handle along the slider to manually control the sequence. The gradual growth in the physical characteristics of players is clearly apparent through the resulting effect. Sequencing: In contrast to animated sequences of the same phenomena changing over time, there are other ways in which a more discrete sequenced experience can suit your needs. This commonly exists by letting users navigate through predetermined, different angles of analysis about a subject. As you navigate through the sequence a narrative is constructed. This is a quintessential example of storytelling with data exploring the metaphor of the anecdote: ‘this happened’ and then ‘this happened’…
Figure 7.6 NFL Players: Height & Weight Over Time
393
The project ‘How Americans Die’ (Figure 7.7) offers a journey through many different angles of analysis. Clicking on the series of ‘pagination’ dots and/or the navigation buttons will take you through a pre-prepared sequence of displays to build a narrative about this subject.
Figure 7.7 Excerpt from ‘How Americans Die’
394
Sometimes data exists in only two states: a before and after view. Using normal animated sequences would be ineffective – too sudden and too jumpy – so one popular technique, usually involving two images, employs the altering of the position of a handle along a slider to reveal/fade the respective views. This offers a more graduated sequence between the two states and facilitates comparisons far more effectively as exhibited by the project shown in Figure 7.8. A different example of sequencing – and an increasingly popular trend – is the vertical sequence. This article from the Washington Post (Figure 7.9) profiles the beauty of baseball player Bryce Harper’s swing and uses a very slick series of illustrations to break down four key stages of his swing action. As you scroll down the page it acts like a lenticular print or flip-book animation. Notice also how well judged the styles of the illustrations are.
Figure 7.8 Model Projections of Maximum Air Temperatures Near the Ocean and Land Surface
395
Figure 7.9 Excerpt from ‘A Swing of Beauty’
396
Contributing: So far the features covered modify the criteria of what data is included/excluded, that then help you dive deeper into the data, and move through sequenced views of that data. The final component of ‘data adjustment’ concerns contributing data. Sometimes there are projects that require user input, either for collecting further records to append and save to an original dataset or just for temporary (i.e. not held beyond the moment of usage) participation. Additionally, there may be scope to invite users to modify certain data in order to inform calculations or customise a display. In each case, the events and controls associated with this kind of interaction are designed to achieve one function: input data.
397
The first example ‘How well do you know your area?’ (Figure 7.10) by ONS Digital, employs simple game/quiz dynamics to challenge your knowledge of your local area in the UK. Using the handle to modify the position along the slider you input a quantitative response to the questions posed. Based on your response it then provides feedback revealing the level of accuracy of your estimation.
Figure 7.10 How Well Do You Know Your Area?
In the next project (Figure 7.11), by entering personal details such as your birth date, country and gender into the respective input boxes you learn about your place in the world’s population with some rather sobering details about your past, present and future on this planet.
Figure 7.11 Excerpt from ‘Who Old Are You?’
398
Figure 7.12 shows an excerpt from ‘512 Paths to the White House’. In this project the toggle buttons are used to switch between three categorical data states (unselected, Democratic and Republican) to build up a simulated election outcome based on the user’s predictions for the winners in each of the key swing states. As each winner is selected, only the remaining possible pathways to victory for either candidate are shown.
Inevitably data privacy and intended usage are key issues of concern for any project that involves personal details being contributed, so be careful to handle this with integrity and transparency.
Adjusting the position of the handle along the slider in the Better Life Index project (Figure 7.13) modifies the quantitative data value representing the weighting of importance you would attach to each quality of life topic. In turn, this modifies the vertical positioning of the country flowers based on the recalculated average quality of life.
Figure 7.12 512 Paths to the White House
399
Figure 7.13 OECD Better Life Index
7.2 Features of Interactivity: Presentation Adjustments In contrast to the features of ‘data adjustment’, this second group of interactive features does not manipulate the data but rather lets you configure the presentation of your data in ways that facilitate assistance and enhance the overall experience.
400
Focusing: Whereas the ‘framing’ features outlined previously modified what data would be included and excluded, ‘focus’ features control what data is visually emphasised and, sometimes, how it is emphasised. Applying such filters helps users select the values they wish to bring to the forefront of their attention. This may be through modifying the effect of depth through colour (foreground, mid- ground and background) or a sorting arrangement. The main difference with the framing features is that no data is eliminated from the display but simply relegated in its contrasting prominence or position.
Figure 7.14 Nobel Laureates
The example in Figure 7.14 provides a snapshot of a project which demonstrates the use of a focus filter. It enables users to select a radio button from the list of options to emphasise different cohorts of all Nobel Laureates (as of 2015). As you can see the selections include filters for women, shared winners and those who were still living at the time. The selected Laureates are not coloured differently, rather
401
the unselected values are significantly lightened to create the contrast.
Figure 7.15 Geography of a Recession
The project shown in Figure 7.15 titled ‘Geography of a Recession’ allows users to select a link from the list of filters provided on the left to emphasise different cohorts of counties across the USA. Once again, the selected counties are not coloured differently here, the unselected regions are de-emphasised by washing-out their original shades.
Figure 7.16 How Big Will the UK Population be in 25 Years’ Time?
402
‘Brushing’ data is another technique used to apply focus filters. In this next example (Figure 7.16), looking at the UK Census estimates for 2011, you use the cursor to select a range of marks from within the ‘violin plot’ display in order to view calculated statistics of those chosen values below the chart. The next example (Figure 7.17), portraying the increase or cuts in Workers’ Compensation benefits by US state, demonstrates a technique known as ‘linking’, whereby hovering over a mark in one chart display will then highlight an associated mark in another chart to draw attention to the relationship. In this case, hovering over a state circle in any of the presented ‘grid maps’ highlights the same state in the other two maps to draw your eye to their respective statuses. You might also see this technique combined with a brushing event to choose multiple data marks and then highlight all associations between charts, as also demonstrated in the population ‘violin plot’ in Figure 7.16.
403
Figure 7.17 Excerpt from ‘Workers’ Compensation Reforms by State’
Sorting is another way of emphasising the presentation of data. In Figure 7.18, featuring work by the Thomson Reuters graphics team, ‘ECB bank test results’, you see a tabular display with sorting features that allow you to reorder columns of data by clicking on the column headers. For categorical data this will sort values alphabetically; for quantitative data, by value order. You can also hand-pick individual records from the table to promote them to the top of the display to facilitate easier comparisons through closer proximity.
Linking and brushing are particularly popular approaches used for exploratory data analysis where you might have several chart panels and wish to see how a single record shows up within each display.
Annotating: As you saw in the previous chapter on data representation, certain combinations of marks and attributes may only provide viewers with a sense of the order of magnitude of the values presented. This might be entirely consistent with the intended tone of the project. However, with interactivity, you can at least enable viewers to interact with marks to view more details momentarily.
404
This temporary display is especially useful because most data representations are already so busy that permanently including certain annotated apparatus (like value labels, gridlines, map layers) would overly clutter the display.
Figure 7.18 Excerpt from ‘ECB Bank Test Results’
The example in Figure 7.19, profiles the use of language throughout the history of US Presidents’ State of the Union addresses, using circle sizes to encode the frequency of different word mentions, giving a gist of the overall quantities and how patterns have formed over time. By hovering over each circle you get access to a tooltip dialogue box which reveals annotations such as the exact word-use quantities and extra contextual commentary. One issue to be aware of when creating pop-up tooltips is to ensure the place they appear does not risk obstructing the view of important data in the chart beneath. This can be especially intricate to handle when you have a lot of annotated detail to share. One tactic is to
405
utilise otherwise-empty space on your page display, occupying it with temporary annotated captions only when triggered by a select or hover event from within a chart. Orientating: A different type of interactive annotation comes in the form of orientation devices, helping you to make better sense of your location within a display – where you are or what values you are looking at. Some of these functions naturally supplement features listed in the previous section about ‘data adjustment’ specifically for navigation support.
Figure 7.19 Excerpt from ‘History Through the President’s Words’
This snapshot, again from the ‘How Americans Die’ project (Figure 7.20), dynamically reveals the values of every mark (both x and y values) in this line chart depending on the hover position of the cursor. This effect is reinforced by visual guides extending out to the axes from the current position.
406
Figure 7.20 Excerpt from ‘How Americans Die’
Figure 7.21 Twitter NYC: A Multilingual Social City
Figure 7.21 displays the language of tweets posted over a period of time from the New York City area. Given the density and number of data points, displaying the details of the mapping layer would be quite cluttered, yet this detail would provide useful assistance for judging the location of the data patterns. The effective solution employed lets you access both views by providing an adjustable slider that allows you to modify the transparency of the network of roads to reveal the apparatus of the mapping layer.
Figure 7.22 Killing the Colorado: Explore the Robot River
407
Finally, as mentioned in the previous section, navigating through digital visualisation projects increasingly uses a vertical landscape to unfold a story (some term this ‘scrollytelling’). Navigation is often seamlessly achieved by using the scroll wheel to move up and down through the display. To assist with orientation, especially when you have a limited field of view of a spatial display, a thumbnail image might be used to show your current location within the overall journey to give a sense of progress. The project featured in Figure 7.22 is a great example of the value of this kind of interface, providing a deep exploration of some of the issues impacting on the Colorado River.
7.3 Influencing Factors and Considerations You now have a good sense of the possibilities for incorporating interactive features into your work, so let’s turn to consider the factors that will have most influence on which of these techniques you might need to or choose to apply.
Formulating Your Brief Skills and resources: Interactivity is unquestionably something that
408
many people aspire to create in their visualisation work, but it is something greatly influenced by the skills possessed, the technology you have access to and what they offer. These will be the factors that ultimately shape your ambitions. Remember, even in common desktop tools like Excel and Powerpoint, which may appear more limited on this front, there are ways to incorporate interactive controls (e.g. using VBA in Excel) to offer various adjustment features (e.g. links within Powerpoint slides to create sequences and navigate to other parts of a document). Timescales: It goes without saying that if you have a limited timeframe in which to complete your work, even with extensive technical skills you are going to be rather pushed in undertaking any particularly ambitious interactive solutions. Just because you want to does not mean that you will be able to. Setting: Does the setting in which the visualisation solution will be consumed lend itself to the inclusion of an interactive element to the experience? Will your audience have the time and know-how to take full advantage of multi-interactive features or is it better to look to provide a relatively simpler, singular and more immediate static solution? Format: What will be the intended output format that this project needs to be created for? What device specifications will it need to work across? How adaptable will it need to be? The range and varied characteristics of modern devices present visualisers (or perhaps more appropriately, at this stage, developers) with real challenges. Getting a visualisation to work consistently, flexibly and portably across device types, browsers and screen dimensions (smartphone, tablet, desktop) can be something of a nightmare. Responsive design is concerned with integrating automatic or manually triggered modifications to the arrangement of contents within the display and also the type and extent of interactive features that are on offer. Your aim is to preserve as much continuity in the core experience as possible but also ensure that the same process and outcome of understanding can be offered to your viewers. While the general trend across web design practice is heading towards a mobile-first approach, for web-based data visualisation developments there is still a strong focus on maximising the capabilities of the desktop experience and then maybe compromising, in some way, the richness of the mobile experience. For ProPublica’s work on ‘Losing Ground’ (Figure 7.23), the
409
approach to cross-platform compatibility was based around the rule of thumb ‘smallify or simplify’. Features that worked on ProPublica’s primary platform of the desktop would have to be either simplified to function practically on the smartphone or simply reduced in size. You will see in the pair of contrasting images how the map display is both shrunk and cropped, and the introductory text is stripped back to only include the most essential information.
Figure 7.23 Losing Ground
Other format considerations include whether your solution will be primarily intended for the Web, but will it also need to work in print? The proverb ‘horses for courses’ comes to mind here: solutions need to be created as fit for the format it will be consumed in. The design features that make up an effective interactive project will unlikely translate directly as a static, print version. You might need to pursue two parallel solutions to suit the respective characteristics of each output format. Another illustration of good practice from the ‘History through the Presidents’ words’ (Figure 7.24) includes a novel ‘Download graphic’ function which, when selected, opens up an entirely different static graphic designed to suit a printable, pdf format.
Figure 7.24 Excerpt from ‘History Through the President’s Words’
410
Purpose map: Interactivity does not only come into your thinking when you are seeking to create ‘Exploratory’ experiences. You may also employ interactive features for creating ‘Explanatory’ visualisations, such as portraying analysis across discrete sequenced
411
views or interactively enabling focus filters to emphasise certain characteristics of the data. The general position defined on the purpose map will not singularly define the need for interactivity, rather it will inform the type of interactivity you may seek to incorporate to create the experience you desire. There will also often be scope for an integrated approach whereby you might lead with an explanatory experience based around showing headline insights and then transitions into a more exploratory experience through offering a set of functions to let users interrogate data in more detail.
Working With Data Data examination: As profiled with the functions to facilitate drill- down navigation, one of the keen benefits of interactivity is when you have data that is too big and too broad to show in one view. To repeat, you can only show so much in a single-screen display. Often you will need to slice up views across and within the various hierarchies of your data. One particular way the physical properties of the data will inform your interaction design choices is with animation. To justify an animated display over time, you will need to consider the nature of the change that exists in your data. If your data is not changing much, an animated sequence may simply not prove to be of value. Conversely, if values are rapidly changing in all dimensions, an animated experience will prove chaotic and a form of change blindness will occur. It may be that the intention is indeed to exhibit this chaos, but the value of animated sequences is primarily to help reveal progressive or systematic change rather than random variation. The speed of an animation is also a delicate matter to judge as you seek to avoid the phenomenon of change blindness. Rapid sequences will cause the stimulus of change to be missed; a tedious pace will dampen the stimulus of change and key observations may be lost. The overall duration will, of course, be informed by the range of values in your temporal data variable. There is no right or wrong here, it is something that you will get the best sense of by prototyping and trialling different speeds.
Establishing Your Editorial Thinking
412
Angle, framing and focus: If you have multiple different angles of analysis you wish to portray then these will have to be accommodated within the space allocated. Alternatively, using interactivity, you could provide access to them via sequenced views or menus enabling their selection. The value of incorporating the potential features to achieve this – and the specific range of different options you do wish to facilitate – will be informed by the scope of the decisions you made in the editorial thinking stages. Thinking again about animations, you must consider whether an animated sequence will ultimately convey the clearest answer to an angle of interest about how something has changed over time. This really depends on what it is you want to show: the dynamics of a ‘system’ that changes over time or a comparison between different states over time? The animated project in Figure 7.25 shows the progressive clearing of snow across the streets of New York City during the blizzard of February 2014. The steady and connected fluidity of progress of the snow-clearing is ideally illustrated through the intervals of change across the 24 hours shown.
Figure 7.25 Plow: Streets Cleared of Snow in New York City
413
Sometimes, you might wish to compare one moment directly against another. With animated sequences, there is a reliance on memory to conduct this comparison of change. However, our ability to recall is fleeting at best and weakens the further apart (in time) the basis of the comparison has occurred. Therefore, to facilitate such a comparison you ideally need to juxtapose individual frames within the same view. The most common technique used to achieve this is through small multiples, where you repeat the same representation for each moment in time of interest and present them collectively in the same view, often through a grid layout. This enables far more incisive comparisons, as you can see through ‘The Horse in Motion’ work by Eadward Muyrbidge, which was used to learn about the galloping form of a horse by seeing each stage of the motion through individually framed moments.
‘Generations of masterpieces portray the legs of galloping horses
414
incorrectly. Before stop-gap photography, the complex interaction of horses’ legs simply happened too fast to be accurately apprehended … but in order to see the complex interaction of moving parts, you need the motion.’ [Paraphrasing] Barbara Tversky and Julie Bauer Morrison, taken from Animation: Can it Facilitate?
Figure 7.26 The Horse in Motion
Data Representation Chart type choice: Some charts are inherently visually complex and ideally need interactivity to make them more accessible and readable for the viewer. The bump chart, chord diagram, and Sankey diagram are just a few of the charts that are far more readable and, by extension, usable if they can offer users the means to filter or focus on certain selected components of the display through interactivity.
Trustworthy Design Functional performance: Faith in the reliability, consistency and general performance of a visualisation is something that impacts on the perception of a project as ‘trustworthy’. Does it do what it promises and can I trust the functions that it performs? Projects that
415
involve the collection of user-inputted data will carry extra risk around trust: how will the data be used and stored? You need to alleviate any such concerns upfront.
‘Confusing widgets, complex dialog boxes, hidden operations, incomprehensible displays, or slow response times … may curtail thorough deliberation and introduce errors.’ Jeff Heer and Ben Schneiderman, taken from Interactive Dynamics for Visual Analysis
Accessible Design Useful: Does it add value? Resort to interactivity only when you have exhausted the possibility of an appropriate and effective static solution. Do not underestimate how effective a well-conceived and executed static presentation of data can be. This is not about holding a draconian view about any greater merits offered by static or print work, but instead recognising that the brilliance of interactivity is when it introduces new means of engaging with data that simply could not be achieved in any other way. Unobtrusive: As with all decisions, an interactive project needs to strive for the optimum ease of usability: minimise the friction between the act of engaging with interactive features and the understanding they facilitate. Do not create unnecessary obstacles that stifle sparks of curiosity and the scent of intrigue that stirs within the user. The main watchword here is affordance, making interactive features seamless and either intuitive or at least efficiently understandable. Visual accessibility: To heighten the accessibility levels of your work you may offer different presentations of it. For people with visual impairments you might offer options to magnify the view of your data and all accompanying text. For those with colour deficiencies, as you will learn about shortly, you could offer options to apply alternative, colour-blind friendly palettes. A further example of this is seen with satellite navigation devices whereby the displayed colour combinations change to better suit the surrounding lightness or darkness at a given time of day.
Elegant Design
416
Feature creep: The discipline required to avoid feature creep is indisputable. The gratuitous interactive operation of today is the equivalent of the flashy, overbearing web design trends of the late 1990s and early 2000s. People were so quick and so keen to show how competent and expressive they could be through this (relatively) new technology that they forgot to judge if it added value. If your audience is quite broad you may be (appropriately) inclined to cover more combinations of features than are necessary in the hope of responding to as many of the anticipated enquiries as well as possible and serving the different types of viewer. Judging the degree of flexibility is something of a balancing act within a single project: you do not want to overwhelm the user with more adjustments than they need, nor do you want to narrow the scope of their likely interrogations. For a one-off project you have to form your own best judgement; for repeatedly used projects you might have scope to accommodate feedback and iteration. Minimise the clicks: With visualisation you are aiming to make the invisible (insights) visible. Conversely, to achieve elegance in design you should be seeking to make visible design features as seamlessly inconspicuous as possible. As Edward Tufte stated, ‘the best design is invisible; the viewer should not see your design. They should only see your content’. Fun: A final alternative influence is to allow yourself room for at least a little bit of fun. So long as the choices do not gratuitously interrupt the primary objective of facilitating understanding, one should not downplay the heightened pleasure that can be generated by interactive features that might incorporate an essence of playability.
Summary: Interactivity Data adjustments affect what data is displayed and may include the following features:
Framing: isolate, include or exclude data. Navigating: expand or explore greater levels of detail in the displayed data. Animating: portray temporal data via animated sequences. Sequencing: navigate through discrete sequences of different angles of analysis.
417
Contributing: customising experiences through user-inputted data.
Presentation adjustments affect how the data is displayed and may include the following features:
Focusing: control what data is visually emphasised. Annotating: interact with marks to bring up more detail. Orientating: make better sense of your location within a display.
Influencing Factors and Considerations
Formulating the brief: skills and resources, timescales, setting, and format will all influence the scope of interactivity. What experience are you facilitating and how might interactive options help achieve this? Working with data: what range of data do you wish to include? Large datasets with diverse values may need interactive features to help users filter views and interrogate the contents. Establishing your editorial thinking: choices made about your chosen angle, as well as definitions for framing and focus will all influence interactive choices, especially if users must navigate to view multiple angles of analysis or representations portrayed through animated sequences. Data representation: certain chart choices may require interactivity to enable readability. Trustworthy design: functional performance and reliability will substantiate the perception of trust from your users. Accessible design: any interactive feature should prove to be useful and unobtrusive. Interactivity can also assist with challenges around visual accessibility. Elegant design: beware of feature creep, minimise the clicks, but embrace the pleasure of playability.
Tips and Tactics
Initial sketching of concepts will be worth doing first before investing too much time jumping into prototype mode. Project management is critical when considering the impact of development of an interactive solution. Backups, contingencies, version control.
418
Do not be precious about – nor overly impressed with – ‘cool’- sounding interaction features that will disproportionately divert precious resources (time, effort, people). Beware of feature creep: keep focusing on what is important and relevant. A technical achievement is great for you, but is it great for the project? Version control and file management will be important here.
419
8 Annotation
Annotation is the third layer of the visualisation design anatomy and is concerned with the simple need to explain things: what is the right amount and type of help your viewers will need when consuming the visualisation?
Annotation is unquestionably the most often neglected layer of the visualisation anatomy. Maybe this is because it involves the least amount of pure design thinking relative to the other matters requiring attention, like interactivity and colour. More likely, it is because effective annotation requires visualisers truly to understand their intended audience. This can be a hard frame of mind to adopt, especially when your potential viewers are likely to have a diverse knowledge, range of interests and capability.
In contrast to the greater theoretical and technical concerns around data representation, colour and interactivity, I find thinking about annotation relatively refreshing. It is not only uncomplicated and based on a huge dose of common sense, but also hugely influential, especially in directly facilitating understanding.
Annotation choices often conform to the Goldilocks principle: too much and the display becomes cluttered, overwhelming, and potentially unnecessarily patronising; too little and the viewers may be inappropriately faced with the prospect of having to find their own way around a visualisation and form their own understanding about what it is showing.
Later in this chapter we will look at the factors that will influence your decision making but to begin with here is a profile of some of the key features of annotated design that exist across two main groups:
Project annotations: helping viewers understand what the project is about and how to use it. Chart annotations: helping viewers perceive the charts and optimise their potential interpretations.
420
8.1 Features of Annotation: Project Annotation This collection of annotation options is related to decisions about how much and what type of help you might need to offer your audiences in their understanding of the background, function and purpose of your project.
Headings: The titles and subtitles occupy such prime real estate within your project’s layout, yet more often than not visualisers fail to exploit these to best effect. There are no universal practices for what a heading should do or say; this will vary considerably between subject areas and industries, but should prove fundamentally useful.
Figure 8.1 A Sample of Project Titles
The primary aim of a title (and often subtitle combination) is to inform viewers about the immediate topic or display, giving them a fair idea about what they are about to see. You might choose to articulate the essence of the curiosity that has driven the project by framing it around a question or maybe a key finding you unearthed following the work. Subheadings, section headings and chart titles will tend to be more functional in their role, making clear to the viewer the contents or focus of attention associated with each component of the display. Your judgement surrounds the level of detail and the type of language
421
you use in each case to fit cohesively with the overall tone of the work. Introductions: Essentially working in conjunction with titles, introductions typically exist as short paragraphs that explain, more explicitly than a title can, what the project is about. The content of this introduction might usefully explain in clear language terms some of the components you considered during the editorial thinking activity, such as:
details of the reason for the project (source curiosity); an explanation of the relevance of this analysis; a description of the analysis (angle, framing) that is presented; expression of the main message or finding that the work is about to reveal (possibly focus).
Some introductions will extend beyond a basic description of the project to include thorough details of where the data comes from and how it has been prepared and treated in advance of its analysis (including any assumptions, modifications or potential shortcomings). There may also be further links to ‘read more’ detail or related articles about the subject.
Figure 8.2 Excerpt from ‘The Color of Debt’
Introductions may be presented as fixed text located near the top (or start) of a project (usually underneath a title) as in Figure 8.2 or, through interactivity, may be hidden from view and brought up in a separate window or pop-up to provide the details upon request. User guides: As you have seen, some projects can incorporate many different features of interactivity. While they may not necessarily be overly technical – and therefore not that hard to learn how to use
422
them – the full repertoire of features may be worth walking through, as in Figure 8.3. This is important to consider so that, as a visualiser, you can be sure your users are acquainted with the entire array of options they have to explore, interrogate and control their experience. You should want people to fully utilise all the different features you have carefully curated and created, so it is in everyone’s interest to think about including these types of user guides.
Figure 8.3 Excerpt from ‘Kindred Britain’
423
424
Multimedia: There is increasing potential and usage of broader media assets in visualisation design work beyond charts, such as video and imagery. In visualisation this is perhaps a relatively contemporary trend (infographics have incorporated such media but visualisations generally have done so far less) and, in some ways, reflects the ongoing blurring of boundaries between this and other related fields. Incorporating good-quality and sympathetically styled assets like illustrations or photo-imagery can be a valuable complementary device alongside your data representation elements. In the ‘Color of Debt’ (Figure 8.4) project, different neighbourhoods of Chicago that have been hardest hit by debt are profiled using accompanying imagery to show more graphic context of the communities affected, including a detailed reference map of the area and an animated panel displaying a sequence of street view images. Imagery, in particular, will be an interesting option to consider when it adds value to help exhibit the subject matter in tangible form, offering an appealing visual hook to draw people in or simply to aid immediate recognition of the topic. In Bloomberg’s billionaires project (Figure 8.5), each billionaire is represented by a pen-and-ink caricature. This is elegant in choice and also dodges the likely flaws of having to compose the work around individual headshot photographs that would have been hard to frame and colour consistently.
Figure 8.4 Excerpt from ‘The Color of Debt’
It was worth Bloomberg investing in the time/cost involved in commissioning these illustrations, given that the project was not a one-off but something that would be an ongoing, updated daily
425
resource. Problems with the integration of such media within a visualisation project will occur when unsuitable attempts are made to combine imagery within the framework of a chart. Often the lack of cohesion creates a significant hindrance whereby the data representations are obscured or generally made harder to read, as the inherent form and colour clashes undermine the functional harmony. Researching, curating, capturing or creating assets of imagery requires skill and a professional approach, otherwise the resulting effect will look amateurish. Incorporating these media into a data visualisation is not about quickly conducting some Google Image fishing exercise. Determining what imagery you will be able to use involves careful considerations around image suitability, quality and, critically, usage rights. Beware the client or colleague who thinks otherwise.
‘Although all our projects are very much data driven, visualisation is only part of the products and solutions we create. This day and age provides us with amazing opportunities to combine video, animation, visualisation, sound and interactivity. Why not make full use of this? … Judging whether to include something or not is all about editing: asking “is it really necessary?”. There is always an aspect of “gut feel” or “instinct” mixed with continuous doubt that drives me in these cases.’ Thomas Clever, Co-founder CLEVER°FRANKE, a data driven experiences studio
A frequent simple example of incorporated imagery is when you have to include logos according to the needs of the organisation for whom your work is being created. Remember to consider this early so you at least know in advance that you will have to assign some space to accommodate this component elegantly.
Footnotes: Often the final visible feature of your display, footnotes provide a convenient place to share important details that further substantiate the explanation of your work. Sometimes this information might be stored within the introduction component (especially if that is interactively hidden/revealed to allow it more room to accommodate detail):
Data sources should be provided, ideally in close proximity to the relevant charts.
426
Credits will list the authors and main contributors of the work, often including the provision of contact details.
Figure 8.5 Excerpt from ‘Bloomberg Billionaires’
Attribution is also important if you wish to recognise the influence of other people’s work in shaping your ideas or to acknowledge the benefits of using an open source application or free typeface, for example. Usage information might explain the circumstances in which the work can be viewed or reused, whether there are any confidentialities or copyrights involved. Time/date stamps are often forgotten but they will give an indication to viewers of the moment of production and from that they might be able to ascertain the work’s current accuracy and contextualise their interpretations accordingly.
Figure 8.6 Excerpt from ‘Gender Pay Gap US’
8.2 Features of Annotation: Chart Annotation This second group of annotated features concerns the ways you provide viewers with specific assistance for perceiving and interpreting the charts. Think of these as being the features that refer directly to your charts or exist directly within or in immediate proximity to each chart.
Reading guides: These are written or visual instructions that provide
427
viewers with a guide for how to read the chart or graphic and offer greater detailed assistance than a legend (see later). The idea of learnability in visualisation is an important consideration. It is a two- way commitment requiring will and effort from the viewer and sufficient assistance from the visualiser. This is something to be discussed in Chapter 11 under ‘Visualisation literacy’. Recognising that their readership may not necessarily understand connected scatter plots, Bloomberg’s visual data team offer a ‘How to Read this Graphic’ guide immediately as you land on the project shown in Figure 8.7. This can be closed but a permanent ‘How to’ button remains for those who may need to refer to it again. The connected scatter plot was the right choice for this angle of analysis, so rather than use a different ‘safer’ representation approach (and therefore alter what analysis was shown) it is to their credit that they respected the capacity of their viewers to be willing to learn how to read this unfamiliar graphical form.
Figure 8.7 Excerpt from ‘Holdouts Find Cheapest Super Bowl Tickets Late in the Game’
428
Figure 8.8 Excerpt from ‘The Life Cycle of Ideas’
429
The second example shown (Figure 8.8) is from the ‘How to Read’ guide taken from the ‘Life Cycle of Ideas’ graphic created by Accurate, a studio renowned for innovative and expressive visualisation work. Given the relative complexity of the encodings used in this piece, it is necessary to equip the viewer with as much guidance as possible to ensure its potential is fully realised. Chart apparatus: Options for chart apparatus relate to the structural components found in different chart types. Every visualisation displayed in this book has different elements of chart apparatus (Figure 8.9), specifically visible axis lines, tick marks or gridlines to help viewers orient their judgements of size and position. There is no right or wrong for including or excluding these features, it tends to be informed by your tonal definitions based on how much precision in the perceiving of values you wish to facilitate. I will discuss the range of different structures underlying each chart type (such as Cartesian, Radial or Spatial) in Chapter 10 on composition, as these have more
430
to do with issues of shape and dimension.
‘Labelling is the black magic of data visualization.’ Gregor Aisch, Graphics Editor, The New York Times
Labels: There are three main labelling devices you will need to think about using within your chart: axis titles, axis labels and value labels:
Axis titles describe what values are being referenced by each axis. This might be a single word or a short sentence depending on what best fits the needs of your viewers. Often the role of an axis is already explained (or implied) by project annotations elsewhere, such as titles or sub-headings, but do not always assume this will be instantly clear to your viewers. Axis labels provide value references along each axis to help identify the categorical value or the date/quantitative value associated with that scale position. For categorical axes (as seen in bar charts and heat maps, for example) one of the main judgments relates to the orientation of the label: you will need to find sufficient room to fit the label but also preserve its readability. For non-categorical data the main judgement will be what scale intervals to use. This has to be a combination of what is most useful for referencing values by your viewer, what is the most relevant interval based on the nature of the data (e.g. maybe a year-level label is more relevant than marking each month), and also what feels like it achieves the best-looking visual rhythm along the chart’s edge. This will be another matter that is discussed more in the composition chapter.
Figure 8.9 Mizzou’s Racial Gap Is Typical On College Campuses
431
Value labels will appear in proximity to specific mark encodings inside the chart. Typically, these labels will be used to reveal a quantity, such as showing the percentage sizes of the sectors in a pie chart or the height of bars. Judging whether to include such annotations will refer back to your definition of the appropriate tone: will viewers need to read off exact values or will their perceived estimates of size and/or relationship be sufficient? The need to include categorical labels will be a concern for maps (whether to label locations?) or charts like the scatter plot seen in Figure 8.9, where you may wish to draw focus to a select sample of the categories plotted across the display.
As you have seen, one way of providing detailed value labels is
432
through interaction, maybe offering a pop-up/tooltip annotation that is triggered by a hover or click event on different mark encodings. Having the option for interactivity here is especially useful as it enables you to reduce clutter from your display that can develop as more annotated detail is added.
Redundancy in labelling occurs when you include value labelling of quantities for all marks whilst also including axis-scale labelling. You are effectively unnecessarily doubling the assistance being offered and so, ideally, you should choose to include one or the other.
Legend: A legend is an annotated feature within or alongside your chart that presents one or several keys to help viewers understand the categorical or quantitative meaning of different attributes.
Figure 8.10 Excerpt from ‘The Infographic History of the World’
For quantitative data the main role for a legend will be if the attribute of area size has been used to encode values, as found on the bubble plot chart type. The keys displayed there will provide a reference for the different size scales. Which selection of sizes to show needs careful thought: what is the most useful guide to help your viewers make their perceptual judgements from a chart? This might not entail showing only even interval sizes (50, 100, 150 etc.); rather, you might offer viewers a indicative spread of sizes to best represent the distribution of your data values. The example in Figure 8.10 shows logical interval sizes to reflect the range of values in the data and also helpfully includes reference to the maximum value size to explain that no shape will be any larger than this. For categorical data you also see a key showing the meaning of different colours and shapes
433
and their associated values.
Figure 8.11 Twitter NYC: A Multilingual Social City
A nice approach to getting more out of your legends is shown in Figure 8.11. Here you will see a key explaining the colour associations combined with a bar chart to display the distribution of quantities for each language grouping from this analysis of tweets posted around New York City. Captions: These exist typically as small passages of written analysis that bring to the surface some of the main insights and conclusions from the work. These might be presented close to related values inside the chart or in separate panels to provide commentary outside the chart. In ‘Gun Deaths’ (Figure 8.12), there is a nice solution that combines annotated captions with interactive data adjustments. Below the main chart there is a ‘What This Data Reveals’ section which some of the main findings from the analysis of the gun death data. The captions double up as clickable shortcuts so that you can quickly apply the relevant framing filters and update the main display to see what the captions are referring to.
Figure 8.12 Excerpt from ‘US Gun Deaths’
434
As creative tools become more ubiquitous the possibility for incorporating non-visual data in you work increases. As an alternative to the written caption there is greater scope to consider using audio as a means of verbally narrating a subject and explaining key messages. Over the past few years one of the standout projects using this feature was the video profiling ‘Wealth Inequality in America’ (Figure 8.13), as introduced in Chapter 3, where the voiceover provides a very compelling and cohesive narrative against the backdrop of the animated visuals that present the data being described.
Figure 8.13 Image taken from ‘Wealth Inequality in America’
435
8.3 Typography As you have seen, many features of annotation utilise text. This means your choices will be concerned not just with what text to include, but also with how it will look. This naturally merits a brief discussion about how typography will have a significant role in the presentation of your work.
Firstly, some clarity about language. A typeface is a designed collection of glyphs representing individual letters, numbers and other symbols of language based on a cohesive style. A font is the variation across several physical dimensions of the typeface, such as weight, size, condensation and italicisation. A typeface can have one or many different fonts in its family. Type effectively represents the collective appearance formed by the choice of typeface and the font.
Tahoma and Century Gothic are different typefaces. This font and this font both belong to the Georgia typeface family but display variations in size, weight and italicisation.
I discussed earlier the distinction between definitions of data visualisation and other related fields. I mentioned how the person creating their design is not necessarily conscious or concerned about what label is attached to their work, they are simply doing their work regardless. The same could be applied to people’s interchangeable use of and meaning of the terms typeface and font, the clarity of which has been irreparably confused by Microsoft’s desktop tools in particular.
Serif typefaces add an extra little flourish in the form of a small line at the end of the stroke in a letter or symbol. Garamond is an example of a serif font. Serif typefaces are generally considered to be easier to read for long sequences of text (such as the full body text) and are especially used in print displays.
Sans-serif typefaces have no extra line extending the stroke for each character. Verdana is an example of a sans-serif typeface. These typefaces are commonly used for shorter sections of text, such as axis or value labels or titles, and for screen displays.
In making choices about which type to use, there are echoes with the thinking you are about to face on using colour. As you will see, colour
436
decisions concern legibility and meaning first, decoration last. With typeface choices you are not dressing up your text, you are optimising its readability and meaning across your display. The desired style of typeface only comes into your thinking after legibility and meaning.
In terms of legibility, you need to choose a typeface and font combination that will be suitable for the role of each element of text you are using. Viewers need to be able to read the words and numbers on display without difficulty. Quite obvious, really. Some typefaces (and specifically fonts) are more easily read than others. Some work better to make numbers as clearly readable as possible, others work better for words. There are plenty of typefaces that might look cool and contemporary but if they make text indecipherable then that is plain wrong.
Typeface decisions will often be taken out of your hands by the visual identity guidelines of organisations and publications, as well as by technical reasons relating to browser type, software compatibility and availability.
Just as variation in colour implies meaning, so does variation in typeface and font. If you make some text capitalised, large and bold-weight this will suggest it carries greater significance and portrays a higher prominence across the object hierarchy than any text presented in lower case, with a smaller size and thinner weight. So you should seek to limit the variation in font where possible.
Text-based annotations should be considered part of the supporting cast and the way you consider typeface and font choices should reflect this role. Typography in visualisation should be seen but really not heard. Deciding on the most suitable type is something that can ultimately come down to experience and influence through exposure to other work. Every individual has their own relied-upon preferences. In practice, I find there is a good chunk of trial and error as well as viewer testing that goes into resolving the final selection. Across the spectrum of data visualisation work being produced there are no significant trends to be informed by largely because judging the most suitable typography choices will be unique to the circumstances influencing each project.
Typography is just another of the many individual ingredients relevant to data visualisation that exists as a significant subject in its own right. It is
437
somewhat inadequate to allocate barely two pages of this book to discussing its role in visualisation, but these will at least offer you a bite- sized window into the topic.
‘Never choose Times New Roman or Arial, as those fonts are favored only by the apathetic and sloppy. Not by typographers. Not by you.’ Matthew Butterick, Typographer, Lawyer and Writer
8.4 Influencing Factors and Considerations Having become familiar with the principal options for annotating, you now have to decide which features to incorporate into your work and how you might deploy these.
Formulating Your Brief
‘Think of the reader – a specific reader, like a friend who’s curious but a novice to the subject and to data-viz – when designing the graphic. That helps. And I rely pretty heavily on that introductory text that runs with each graphic – about 100 words, usually, that should give the new-to- the-subject reader enough background to understand why this graphic is worth engaging with, and sets them up to understand and contextualize the takeaway. And annotate the graphic itself. If there’s a particular point you want the reader to understand, make it! Explicitly! I often run a few captions typeset right on the viz, with lines that connect them to key elements in the design.’ Katie Peek, Data Visualisation Designer and Science Journalist, on making complex and/or complicated subject matter accessible and interesting to her audience
Audience: Given that most annotations serve the purpose of viewer assistance, your approach will inevitably be influenced by the characteristics of your intended audience. Having an appreciation of and empathy towards the knowledge and capabilities of the different cohorts of viewers is especially important with this layer of design. How much help will they need to understand the project and also the data being portrayed? You will need to consider the following:
Subject: how well acquainted will they be with this subject matter? Will they understand the terminology, acronyms,
438
abbreviations? Will they recognise the relevance of this particular angle of analysis about this subject? Interactive functions: how sophisticated are they likely to be in terms of being able to understand and utilise the different features of interactivity made possible through your design? Perceiving: how well equipped are they to work with this visualisation? Is it likely that the chart type(s) will be familiar or unfamiliar; if the latter, will they need support to guide them through the process of perceiving? Interpreting: will they have the knowledge required to form legitimate interpretations of this work? Will they know how to understand what is good or bad, what big and small mean, what is important, or not? Alternatively, will you need to provide some level of assistance to address this potential gap?
Purpose map: The defined intentions for the tone and experience of your work will influence the type and extent of annotation features required. If you are working towards a solution that leans more towards the ‘reading’ tone you are placing an emphasis on the perceptibility of the data values. It therefore makes sense that you should aim to provide as much assistance as possible (especially through extensive chart annotations) to maximise the efficiency and precision of this process. If it is more about a ‘feeling’ tone then you may be able to justify the absence of the same annotations. Your intent may be to provide more of a general sense – a ‘gist’ – of the order of magnitude of values. If you are seeking to provide an ‘explanatory’ experience it would be logical to employ as many devices as possible that will help inform your viewers about how to read the charts (assisting with the ‘perceiving’ stage of understanding) and also bring some of the key insights to the surface, making clear the meaning of the quantities and relationships displayed (thus assisting with the stage of ‘interpreting’). The use of captions and visual overlays will be particularly helpful in achieving this, as will the potential for audio accompaniments if you are seeking to push the explanatory experience a step further. ‘Exploratory’ experiences are less likely to include layers of insight assistance, instead the focus will be more towards project-level annotation, ensuring that viewers (and particularly here, users) have as much understanding as possible about how to use the project for their exploratory benefit. You might find, however, that devices like
439
‘How to read this graphic’ are still relevant irrespective of the definition of your intended experience. Characteristically, ‘exhibitory’ work demonstrates far less annotated assistance because, by intention, it is more about providing a visual display of the data rather than offering an explanatory presentation or the means for exploratory interrogation. The assumptions here are that audiences will have sufficient domain and project knowledge not to require extensive additional assistance. Common chart annotations like value labels and legends, and project annotations like titles and introductions, are still likely to be necessary, but these might reflect the extent required.
Establishing Your Editorial Thinking Focus: During your editorial thinking you considered focus and its particular role in supporting explanatory thinking. Are there specific value labels that you wish to display over others? Rather than labelling all values, for example, have you determined that only certain marks and attributes will merit labelling? As you saw earlier in the example scatter plot about the under-representation of black students in US colleges, only certain points were labelled, not all. These would have been judged to have been the most relevant and interesting elements to emphasise through annotation.
Trustworthy Design Transparency: Annotation is one of the most important aids to ensure that you secure and sustain trust from your viewers by demonstrating integrity and openness:
Explain what the project is and is not going to show. Detail where the data came from and what framing criteria were used during the process of acquisition, and also make what has been ultimately included in the chart(s). Outline any data transformation treatments, assumptions and calculations. Are there any limitations that viewers need to be aware of? Highlight and contextualise any findings to ensure accuracy in interpretation. With digital projects in particular, provide access to coding
440
repositories to lay open all routines and programmatic solutions.
Accessible Design Understandable: If you recall, in the section profiling circumstances you considered what the characteristics were of the setting or situation in which your audience might consume your visualisation. Well-judged project and chart annotations are entirely concerned with providing a sufficient level of assistance to achieve understanding. The key word there is ‘sufficient’ because there is a balance: too much assistance makes the annotations included feel overburdening; too little and there is far more room for wrong assumptions and misconceptions to prosper. A setting that is consistent with the need to deliver immediate insights will need suitable annotations to fulfil this. There will be no time or patience for long introductions or explanations in that setting. Conversely, a visualisation about a subject matter that is inherently complex may warrant such assistance.
Elegant Design Minimise the clutter: A key concern about annotations is judging the merits of including structural or textual assistance against the potential disruption and obstruction caused by these to the view of the data. Any annotation device added to your display has a spatial and visual consequence that needs to be accommodated. Of course, as mentioned, with the benefit of interactivity it is possible to show and hide layers of detail. Overall, you will have to find the most elegant solution for presenting your annotations to ensure you do not inadvertently undermine the help you are trying to provide.
Summary: Annotation Project annotations help viewers understand what the project is about and how to use it, and may include the following features:
Headings: titles, sub-titles and section headings. Introductions: providing background and aims of the project.
441
User guides: advice or instruction for how to use any interactive features. Multimedia: the potential to enhance your project using appropriate imagery, videos or illustrations. Footnotes: potentially includes data sources, credits, usage information, and time/date stamps.
Chart annotations help viewers perceive the charts and optimise their potential interpretations and may include the following features:
Chart apparatus: axis lines, gridlines, tick marks. Labels: axis titles, axis labels, value labels. Legend: providing detailed keys for colour or size associations. Reading guides: detailed instructions advising readers how to perceive and interpret the chart. Captions: drawing out key findings and commentaries.
Typography Most of the annotation features you include are based on text and so you will need to consider carefully the legibility of the typeface you choose and the logic behind the font-size hierarchy you display.
Influencing Factors and Considerations
Formulating the brief: consider the characteristics and needs of the audience. Certain chart choices and subjects may require more explanation. From the ‘purpose map’ what type of tone and experience are you trying to create and what role might annotation play? Establishing your editorial thinking: what things do you want to emphasise or direct the eye towards (focus)? Trustworthy design: maximise the information viewers have to ensure all your data work is transparent and clearly explained. Accessible design: what is the right amount and type of annotation suitable to the setting and complexity of your subject? Elegant design: minimise the clutter.
Tips and Tactics
Attention to detail is imperative: all instructions, project information, captions and value labels need to be accurate. Always spell-check
442
digitally and manually, and ask others to proofread if you are too ‘close’ to see. Do not forget to check on permission to use any annotated asset, such as imagery, photos, videos, quotations, etc.
443
9 Colour
Having established which charts you will use, the potential interactive functions that might be required and the annotation features that will be especially useful, you have effectively determined all the visible elements that will be included in your project. The final two layers of design concern not what elements will be included or excluded, but how they will appear. After this chapter you will look at issues on composition, but before that the rather weighty matter of colour.
As one of the most powerful sensory cues, colour is a highly influential visual property. It is arguably the design decision that has the most immediate impact on the eye of the viewer. All the design features of your visualisation display hold some attribute of colour, otherwise they are invisible:
Every mark and item of apparatus in your charts will be coloured; indeed colour in itself may be an attribute that represents your data values. Interactive features do not always have an associated visible property (some are indeed invisible and left as intuitively discoverable). However, those features that involve buttons, menus, navigation tabs and value sliders will always have a colour. Annotation properties such as titles, captions and value labels will all be coloured. Composition design mainly concerns the arrangement of all the above features, though you might use colour to help achieve a certain design layout. As you will see, emptiness is a useful organising device – leaving something blank is a colour choice.
Thankfully, there is a route through all of this potential complexity relying on just a little bit of science mixed in with lots of common sense. By replacing any arbitrary judgements that might have been previously based on taste, and through increasing the sensitivity of your choices, colour becomes one of the layers of visualisation design that can be most quickly and significantly improved.
444
‘Colors are perhaps the visual property that people most often misuse in visualization without being aware of it.’ Robert Kosara, Senior Research Scientist at Tableau Software
The key factor in thinking about colour is to ensure you establish meaning first and decoration last. That is not to rule out the value of certain decorative benefits of colour, but to advise that these should be your last concern. Besides, in dealing with meaningful applications of colour you will already have gone a long way towards establishing the ‘decorative’ qualities of your project’s aesthetic appearance.
This chapter begins with a look at some of the key components of colour science, offering a foundation for your understanding about this topic. After that you will learn about the ways and places in which colour could be used. Finally, you will consider the main factors that influence colour decisions.
COLOUR thinking begins from inside the chart(s), working outwards across the rest of the visualisation anatomy:
Data legibility. Editorial salience. Functional harmony.
9.1 Overview of Colour Theory COLOUR in visualisation is something of a minefield. As with many of these design layer chapters, an introduction to colour involves judging the right amount of science and the right amount of practical application. What does justice to the essence of the subject and gives you the most relevant content to work with is a delicate balance.
When you lift the lid on the science behind colour you open up a world of brain-ache. When this chapter is finalised I will have spent a great deal of time agonising over how to explain this subject and what to leave in or leave out because there is so much going on with colour. And it is tricky. Why? Because you almost come face to face with philosophical questions like ‘what is white?’ and the sort of mathematical formulae that you really rather hoped had been left behind at school. You learn how the colours you
445
specify in your designs as X might be perceived by some people as Y and others as Z. You discover that you are not just selecting colours from a neat linear palette but rather from a multi-dimensioned colour space occupying a cubic, cylindrical or spherical conceptual shape, depending on different definitions.
The basis of this topic is the science of optics – the branch of physics concerned with the behaviour and properties of light – as well as colorimetry – the science and technology used to quantify and describe human colour perception. Two sciences, lots of maths, loads of variables, endless potential for optical illusions and impairment: that is why colour is tricky and why you need to begin this stage of thinking with an appreciation of some colour theory.
The most relevant starting point is to recognise that when dealing with issues of colour in data visualisation you will almost always be creating work on some kind of computer. Unless you are creating something by hand using paints or colouring pencils, you will be using software viewed through an electronic display.
This is important because a discussion about colour theory needs to be framed around the RGB (Red, Blue, Green) colour model. This is used to define the combination of light that forms the colours you see on a screen, conceptually laid out in a cubic space based on variations across these three attributes.
The output format of your work will vary between screen display and print display. If you are creating something for print you will have to shift your colour output settings to CMYK (Cyan, Magenta, Yellow and Black). This is the model used to define the proportions of inks that make up a printed colour. This is known as a subtractive model, which means that combining all four inks produces black, whereas RGB is additive as the three screen colours combine to produce white.
When you are creating work to be consumed on the Web through screen displays, you will often program using HEX (Hexadecimal) codes to specify the mix of red, green and blue light (in the form #RRGGBB using codes 00 to FF).
While CMYK communicates from your software to a printer, telling it
446
what colours to print as an output, it does not really offer a logical model to think about the input decisions you will make about colour. Neither, for that matter, does RGB: it just is not realistic to think in those terms when considering what choices are needed in a visualisation design. There are different levers to adjust and different effects being sought that require an alternative model of thinking.
Figure 9.1 HSL Colour Cylinder
Figure 9.2 Colour Hue Spectrum
I share the belief with many in the field that the most accessible colour model – in terms of considering the application of colour in data visualisation – is HSL (Hue, Saturation, Lightness), devised by Albert Munsell in the 1980s. These three dimensions combine to make up what is known as a cylindrical-coordinate colour representation of the RGB colour model (I did warn you about the cylinders).
Hue is considered the true colour. With hue there are no shades (adding black), tints (adding whites) or tones (adding grey) – a
447
consideration of these attributes follows next. When you are describing or labelling colours you are most commonly referring to their hue: think of the colours of the rainbow ranging through various mixtures of red, orange, yellow, green, blue, indigo and violet. Hue is considered a qualitative colour attribute because it is defined by difference and not by scale. Saturation defines the purity or colourfulness of a hue. This does have a scale from intense pure colour (high saturation) through increasing tones (adding grey) to the no-colour state of grey (low saturation). In language terms think vivid through to muted.
Figure 9.3 Colour Saturation Spectrum
Lightness defines the contrast of a single hue from dark to light. It is not a measure of brightness – there are other models that define that – rather a scale of light tints (adding white) through to dark shades (adding black). In language terms I actually think of lightness more as degrees of darkness, but that is just a personal mindset.
Figure 9.4 Colour Lightness Spectrum
Technically speaking, black, white and grey are not considered colours.
I have deliberately described these dimensions separately because, as you will see when looking at the applications of colour in visualisation, your decisions will often be defined by how you might employ these distinct dimensions of colour to form your visual display. The main choices tend to fall between employing difference in hue and variation in lightness, with the different levels of saturation often being a by-product of the definitions made for the other two dimensions.
Alternative models exist offering variations on a similar theme, such as HSV (Hue, Saturation, Value), HSI (Hue, Saturation, Intensity), HSB (Hue, Saturation, Brightness) and HCL (Hue, Chroma, Luminance).
These are all primarily representations of the RGB model space but involve differences in the mathematical translation into/from RGB and
448
offer subtle differences in the meaning of the same terms (local definitions of hue and saturation vary). The biggest difference relates to their emphasis as a means of specifying either a colour quality (in an input, created sense) or a colour perception (in how a colour is ultimately experienced).
Pantone is another colour space that you might recognise. It offers a proprietary colour-matching, identifying and communicating service for print, essentially giving ‘names’ to colours based on the CMYK process.
The argument against using the HSL model for defining colour is that, while it is fine for colour setting (i.e. an intuitive way to think about and specify the colours you want to set in your visualisation work), the resulting colours will not be uniformly perceived the same, from one device to the next. This is because there are many variables at play in the projection of light to display colour and the light conditions present in the moment of perception. That means the same perceptual experience will not be guaranteed. It is argued that more rigorous models (such as CIELAB) offer an absolute (as opposed to a relative) definition of colour for both input and output. My view is that they are just a little bit too hard to easily translate into visualisation design thinking. Furthermore, trying to control for all the subtleties of variation in consumption conditions is an extra burden you should ideally avoid.
At this stage, it is important to be pragmatic about colour as much as possible. The vast majority of your colour manipulating and perceptual needs should be nicely covered by the HSL model. As and when you develop a deeper, purist interest in colour you should then seek to learn more about the nuances in the differences between the definitions of these models and their application.
9.2 Features of Colour: Data Legibility Data legibility concerns the use of the attribute of colour to encode data values in charts. The objective here is to make the data being represented by differences in colour as clearly readable and as meaningful as possible.
While you have probably already decided by now the chart or charts you
449
intend to use, you still need to take think carefully – and separately – about how you will specifically employ colour. To do this we first need to revisit the classification of data types and consider how best to use colour for representing each different type.
Nominal (Qualitative) With nominal data colour is used to classify different categorical values. The primary motive for the choice of colour is to create a visible distinction between each unique categorical association, helping the eye to discern the different categories as efficiently and accurately as possible.
Creating contrast is the main aim of representing nominal data. What you are not seeking to show or even imply is any sense of an order of magnitude. You want to help differentiate one category from the next – and make it easily identifiable – but to do so in a way that preserves the sense of equity among the colours deployed.
Figure 9.5 Excerpt from ‘Executive Pay by the Numbers’
Variation in hue is typically the colour dimension to consider using for
450
differentiating categories. Additionally, you might explore different tones (variations in saturation across the hues). You should not, though, consider using variations in the lightness dimension. That is because the result is insufficiently discernible. As you can see demonstrated in Figure 9.5, the lightness variation of a blue tone makes it quite hard to connect the colour scale presented in the key at the top with the colours displayed in the stacked bars underneath. With the shading in the column header and the 2011 grey bar also contributing similar tones to the overall aesthetic of the table our visual processing system has to work much harder to determine the associations than it should need to do.
Often the categories you will be differentiating with colour will be relatively few in number, maybe two or three, such as in the separation between political parties or plotting different values for gender, as seen in Figure 9.6.
Figure 9.6 How Nations Fare in PhDs by Sex
Figure 9.7 How Long Will We Live – And How Well?
451
Beyond these small numbers, you still typically might only need to contend with assigning colours to around four to six categories, perhaps in analysis that needs to visually distinguish values for different continents of the world, as seen in the scatter plot in Figure 9.7.
As the range of different categories grows, the ability to preserve clear differentiation becomes harder. In expanding your required palette, the colours used become decreasingly unique. The general rule of thumb is that once you have more than 12 categories it will not be possible to find a sufficiently different colour to assign to categories from 13 upwards. Additionally, you are really increasing the demands of learning and recognition for viewers. This then becomes quite a cognitive burden and delays the process of understanding.
Figure 9.8 Charting the Beatles: Song Structure
452
Two approaches for dealing with this. Firstly, consider offering interactive filters to modify what categories are displayed in a visualisation – thus potentially reducing the impact of so many being available. Secondly, think about transforming your data by excluding or combining categories in to a reduced number of aggregate groupings.
Depending on the subject of your data, sometimes you can look to supplement the use of colour with texture or pattern to create further visible distinctions. In Figure 9.8 you can see two patterns being used occasionally as additive properties to show the structure of tracks on The Beatles’ album.
Ordinal (Qualitative) With ordinal data you are still dealing with categories but now they have a natural hierarchy or ordering that can be exploited. The primary motive for using colour in this case is not only to create a visible distinction between each unique category association but also to imply some sense of an order of magnitude through the colour variation. The colour dimensions used to achieve this tend to employ variations of either the saturation or the lightness (or a combination of both). You might also introduce different hues when dealing with diverging (dual-direction) scales rather than simply converging (single-direction) ones.
453
Figure 9.9 displays a simple example of colour used to display a converging ordinal variable. This is the teacup that I use in my office. On the inside you can see it has a colour guide to help ascertain how much milk you might need to add: going through Milky, Classic British, Builder’s Brew, and finally Just Tea (zero milk).
Figure 9.9 Photograph of MyCuppa Mug
A typical example of a diverging ordinal scale might be seen in the stacked bar chart showing the results of a survey question (Figure 9.10). The answers are based on the strength of feeling: strongly agree, agree, neutral, disagree, strongly disagree. By colouring the agreement in red (‘hot’ sometimes used to represent ‘good’) and the disagreement in blue (‘cold’
454
to mean ‘bad’) means a viewer can quickly perceive the general balance of feelings being expressed.
Figure 9.10 Example of a Stacked Bar Chart Based on Ordinal Data
Another example of ordinal data might be to represent the notion of recency. In Figure 9.11 you see a display plotting the 2013 Yosemite National Park fire. Colour is used to display the recorded day-by-day progress of the fire’s spread. The colour scale is based on a recency scale with darker = recent, lighter = furthest away (think faded memory).
Figure 9.11 The Extent of Fire in the Sierra Nevada Range and Yosemite National Park, 2013
455
Interval and Ratio (Quantitative) With quantitative data (ratio and interval) your motive, as it is with ordinal data, is to demonstrate the difference between and of a set of values. In the choropleth map in Figure 9.12, showing the variation in electricity prices across Switzerland, the darker shades of blue indicate the higher values, the lighter tints the lower prices. This approach makes the viewer’s perception of the map’s values immediate – it is quite intuitive to recognise the implication of the general patterns of light and dark shades.
Figure 9.12 What are the Current Electricity Prices in Switzerland [Translated]
456
Typically, using colour to represent quantitative data will involve breaking up your data values into discrete classifications or ‘bins’. This makes the task of reading value ranges from their associated colour shade or tone a little easier than when using a continuous gradient scale. While our capacity to judge exact variations in colour is relatively low (even with a colour key for reference), we are very capable of detecting local variations of colour through differences in tint, shade or tone. Assessing the relative contrast between two colours is generally how we construct a quantitative hierarchy.
Look at the fascinating local patterns that emerge in the next map (Figure 9.13), comparing increases in the percentage of people gaining health insurance in the USA (during 2013–14). The data is broken down to county level detail with a colour scale showing a darker red for the higher percentage increases.
Some of the most relevant colour practices for data visualisation come from the field of cartography (as do many of the most passionate colour
457
purists). Just consider the amount of quantitative and categorical detail shown in a reference map that relies on colour to differentiate types of land, indicate the depth of water or the altitude of high ground, present route features of road and rail networks, etc. The best maps pack an incredible amount of detail into a single display and yet somehow they never feel disproportionately overwhelming.
Figure 9.13 Excerpt from ‘Obama’s Health Law: Who Was Helped Most’
Aside from the big-picture observations of the darker shades in the west and the noticeably lighter tints to the east and parts of the mid-west, take a closer look at some of the interesting differences at a more local level. For example, notice the stark contrast across state lines between the dark regions of southern Kentucky (to the left of the annotated caption) and the light regions in the neighbouring counties of northern Tennessee. Despite their spatial proximity there are clearly strong differences in enrolment on the programme amongst residents of these regions.
Both of these previous examples use a convergent colour scale, moving through discrete variations in colour lightness to represent an increasing
458
scale of quantitative values, from zero or small through to large. As illustrated with the stacked bar chart example shown earlier, portraying the range of feelings from an ordinal dataset, sometimes you may need to employ a divergent colour scale. This is when you want to show how values are changing in two directions either side of a set breakpoint.
Figure 9.14 Daily Indego Bike Share Station Usage
Figure 9.14 shows a cropped view of a larger graphic comparing the relative peaks and troughs of usage across all bike share stations in Philadelphia over a 24-hour period. The divergent colour scale uses two hues and variations in lightness to show the increasingly busy and increasingly slow periods of station activity either side of a breakpoint, represented by a very light grey to indicate the average point. The darkest red means the station is full, the darkest blue means the station is empty.
Regardless of whether you are plotting a converging or diverging scale, judging how you might divide up your colour scales into discrete value bins needs careful thought. The most effective colour scales help viewers perceive not just the relative order of magnitude – higher or lower – but also a sense of the absolute magnitude – how different a value might be compared to another value.
There is no universal rule about the number of value bins. Indeed, it is not uncommon to see entirely continuous colour scales. However, a general rule of thumb I use is that somewhere between between four and nine meaningful – and readable – value intervals should suffice. There are two key factors to consider when judging your scales:
Are you plotting observed data or observable data? You might only
459
have collected data for a narrow range of quantities (e.g. 15 to 35) so will your colour classifications be based on this observed range or on the potentially observable data range i.e. the values you know would/could exist with a wider sample size or on a different collection occasion (e.g. 0 to 50)? What are the range and distribution of your data? Does it make sense to create equal intervals in your colour classifications or are there more meaningful intervals that better reflect the shape of your data and the nature of your subject? Sometimes, you will have legitimate outliers that, if included, will stretch your colour scales far beyond the meaningful concentration of most of your data values.
Figure 9.15 Battling Infectious Diseases in the 20th Century: The Impact of Vaccines
You can see this effect in Figure 9.15, showing the incidence of Hepatitis A per 100,000 population. There are only three values that exceed 100 (you can see them on the top line for Alaska in the late 1970s). To accommodate these outliers the colour scale becomes somewhat stretched- out, with a wide range of potential values being represented by a dark yellow to red colour. With 99.9% of the values being under 100 there is little discernibly in the blue/green shades used for the lower values. If outliers are your focus, it makes sense to include these and colour
460
accordingly to emphasise their exceptional quality. Otherwise if they risk compromising the discrete detail of the lower values you might look to create a broad classification that uses a single colour for any value beyond a threshold of maybe 75, with even value intervals of maybe 15 below that help to show the patterns of smaller values.
For diverging scales, the respective quantitative shades either side of a breakpoint need to imply parity in both directions. For example, a shade of colour that means +10% one side of the breakpoint should have an equal shade intensity in a different hue on the other side to indicate the same interval, i.e. −10%. Additionally, the darkest shades of hues at the extreme ends of a diverging scale must still be discernible. Sometimes the darkest shades will be so close to black that you will no longer be able to distinguish the differences in their underlying hues when plotted in a chart or map.
As well as considering the most appropriate discrete bins for your values, for diverging scales one must also pay careful attention to the role of the breakpoint. This is commonly set to separate values visually above or below zero or those either side of a meaningful threshold, such as target, average or median.
One of the most common mistakes in using colour to represent quantitative data comes with use of the much-derided rainbow scale. Look at Figure 9.16, showing the highest temperatures across Australia during the first couple of weeks in 2013. Consider the colour key to the right of the map and ask yourself if this feels like a sufficiently intuitive scale. If the key was not provided, would you be able to perceive the order of magnitude relationship between the colours on the map? If you saw a purple colour next to a blue colour, which would you expect to mean hotter and which colder?
Figure 9.16 Highest Max Temperatures in Australia
461
While the general implication of blue = ‘colder’ through to red = ‘hotter’ is included within sections of this temperature colour scale, it is the presence of many other hues that obstructs the accessibility and creates inconsistency in logic. For instance, do the colours used to show 24°C (light blue) jumping to 26°C (dark green) make sense as a means for showing an increasing temperature? How about 18°C (grey) to 20°C (dark blue), or the choice of the mid-brown used for 46°C which interrupts the increasingly dark red sequence? If you saw on the map a region with the pink tone as used for 16°C would you be confident that you could easily distinguish this from the lighter pink used to represent 38°C? Unless there are meaningful thresholds within your quantitative data – justifiable breakpoints – you should only vary your colour scales through the lightness dimension, not the hue dimension.
One of the interesting recurring challenges faced by visualisers is how to represent nothing. For example, if a zero quantity or no category is a meaningful state to show, you still need to represent this visually somehow, even though it might possess no size, no position and no area. How do you distinguish between no data and a zero value?
Figure 9.17 State of the Polar Bear
462
Typically, using colour is one of the best ways to portray this. Figure 9.17 shows one solution to making ‘no data’ a visible value. This map displays the population trends of the polar bear. Notice those significant areas of grey representing ‘data deficient’. A subtle but quite effective political point is being made here by including this status indicator. As I mentioned before, sometimes the absence of data can be the message itself.
Figure 9.18 Excerpt from ‘Geography of a Recession’
463
When considering colour choices for quantitative classifications, you will need to think especially carefully about the lowest value grouping: is it to be representative of zero, an interval starting from zero up to a low value, or an interval starting only from the minimum value and never including zero? In this choropleth map (Figure 9.18) looking at the unemployment rate across the counties of the USA, no value is as low as zero. There might be value that are close, but nowhere is the unemployment rate at 0%. As you can see, the lowest tint used in this colour key is not white, rather a light shade of orange, so as not to imply zero. Whilst not relevant to this example, if you wanted to create a further distinction between the lowest value interval and the ‘null’ or ‘no data’ state you could achieve this by using a pure white/blank.
9.3 Features of Colour: Editorial Salience Having considered options for the application of colour in facilitating data legibility, the next concern is colour used for editorial salience. Whereas data legibility was concerned with helping to represent data, using colour for editorial salience is about drawing the viewer’s attention to the
464
significant or meaningful features of your display. Colour offers such a potent visual stimulus and an influential means for drawing out key aspects of your data and project that you might feel are sufficiently relevant to make prominent.
Consider again the idea of photography and the effect of taking a photograph of a landscape. You will find the foreground objects are darker and more prominent than the faded view of the background in the distance as light and colour diminish. Using colour to achieve editorial salience involves creating a similar effect of depth across your visualisation’s contents: if everything is shouting, nothing is heard.
The goal of using colour to facilitate editorial salience is a suitable contrast. For things to stand out, you are in turn determining which other things will not.
The degree of contrast you might seek to create will vary. Often you will be seeking to draw a significant contrast, maximising the emphasis of a value or subset of values so the viewer can quickly home in on what you have elevated for their attention relative to everything else.
For this reason, grey will prove to be one of your strongest allies in data visualisation. When contrasted with reasonably saturated hues, grey helps to create depth. Elements coloured in greyscale will sit quietly at the back of the view, helping to provide a deliberately subdued context that enables the more emphasised coloured properties to stand proudly in the foreground.
In Figure 9.19, the angle of analysis shows a summary of the most prevalent men’s names featuring among the CEOs of the S&P 1500 companies. As you can see there are more guys named ‘John’ or ‘David’ than the percentage of all the women CEOs combined. With the emphasis of the analysis on this startling statement of inequality the bar for ‘All women’ is emphasised in a burgundy colour, contrasting with the grey bars of all the men’s names. Notice also that the respective axis and bar value labels are both presented using a bold font, which further accentuates this emphasis. It is also editorially consistent with the overriding enquiry of the article. As discussed in Chapter 3, bringing to the surface key insights from data displays in this way contributes towards facilitating an ‘explanatory’ experience.
465
Figure 9.19 Fewer Women Run Big Companies Than Men Named John
Figure 9.20 NYPD, Council Spar Over More Officers
Sometimes, only noticeable contrast – not shouting, just being slightly more distinguishable – may be appropriate. Compared with the previous bar chart example, Figure 9.20 creates a more subtle distinction between the slightly darker shade of green (and emboldened text) emphasising the New York figures compared to the other listed departments in a slightly lighter green. As with the CEOs’ example, the object of our attention is the
466
subject of focus in the analysis, in this case regarding a drive for more NYPD officers. This does not need to be any more contrasting; it is just as sufficiently noticeable as the visualiser wishes it to be.
Sometime you will seek to create several levels of visual ‘urgency’ in the relative contrast of your display. The colour choices in Figure 9.21 gives foreground prominence to the yellow coloured markers and values (the dots are also larger) and then mid-ground/secondary prominence to the slightly muted red markers. In perceiving the values of the yellow markers, the viewer is encouraged to concentrate on primarily comparing these with the red markers. The subtle grey markers are far less visible – closer in shade to the background than the foreground – and deliberately relegated to a tertiary level so they do not clutter up the display and cause unwarranted attention. They provide further context for the distribution of the values but do not need to be any more prominent in their relationship with the foreground and mid-ground colours.
Figure 9.21 Excerpt from a Football Player Dashboard
I touched on the use of encoded overlays earlier where coloured areas or bandings can be used to help separate different regions of a display in order to facilitate faster interpretation of the meaning of values. In the bubble plot in Figure 9.22, you can see the circle markers are colour coded
467
to help viewers quickly ascertain the significance of each location on the chart according to the quadrants in which they fall. Notice how in the background the diagonal shading further emphasises the distinction between above the line ‘improvement’ and below the line ‘worsening’, a very effective approach.
Figure 9.22 Elections Performance Index
9.4 Features of Colour: Functional Harmony After achieving data legibility and editorial salience through astute colour choices, functional harmony is concerned with ensuring that any remaining colour choices will aid, and not hinder, the functional effectiveness and elegance of the overall visualisation.
‘When something is not harmonious, it’s either boring or chaotic. At one extreme is a visual experience that is so bland that the viewer is not engaged. The human brain will reject under-stimulating information. At the other extreme is a visual experience that is so overdone, so chaotic, that the viewer can’t stand to look at it. The human brain rejects what it can not organise, what it cannot understand.’ Jill Morton, Colour
468
Expert and Researcher
You must judge the overall balance of and suitability of your collective colour choices and not just see these as isolated selections. This is again primarily a judgement about contrast – what needs to be prominent and what needs to be less so. Such an apparent calming quality about a well- judged and cohesive colour palette is demonstrated by Stefanie Posavec’s choices in visualising the structure of Walter Benjamin’s essay ‘Art in the age of mechanical reproduction’ (Figure 9.23). There is effortless harmony here between the colour choices extending across the entire anatomy of design: the petals, branches, labels, titles, legend, and background.
A reminder that any and every design feature you incorporate into your display will have a property of colour otherwise they will be invisible. In looking at data legibility and editorial salience you have considered your colour choices for representing data. A desire to achieve functional harmony means considering further colour decisions that will help establish visual relationships across and between the rest of your visualisation’s anatomy: its interactive features, annotations and composition.
Figure 9.23 Art in the Age of Mechanical Reproduction: Walter Benjamin
Interactive features: Visible interactive features will include controls such as dropdown menus, navigation buttons, time sliders and parameter selectors. The colour of every control used will need to be harmonious with the rest of the project but also, critically, must be functionally clear. How you use colour to help the user discern what
469
is selected and what is not will need to be carefully judged. To illustrate this, Figure 9.24 shows an interactive project that examines the connected stories of the casualties and fatalities from the Iraqi and Afghan conflicts. Here you can see that there are several interactive features, all of which are astutely coloured in a way that feels both consistent with the overall tone of the project but also makes it functionality evident what each control’s selected status or defined setting is. This is achieved through very subtle but effective combinations of dark and light greys that help create intuitive clarity about which values the user has selected or highlighted. When a button has a toggle setting (on/off, something/something else), such as the ‘Afghanistan’ or ‘Iraq’ tabs at the top, the selected tab is highlighted in bright grey and the unselected tab in a more subdued grey. Filters can either frame (include/exclude) or focus (highlight/relegate) the data. The same approach to using brighter greys for the selected parameter values makes it very clear what you have chosen, but also what you have excluded (while making evident the other currently-unselected values from which you can potentially choose).
Figure 9.24 Casualties
Annotations: Chart annotations such as gridlines, axis lines and value labels all need colouring in a way that will be sympathetic to the colour choices already made for the data representation and, possibly, editorial contrasting. As mentioned in the last chapter, many
470
annotation devices exist in the form of text and so the relative font colour choices will need to be carefully considered. For any annotation device the key guiding decision is to find the level at which these are suitably prominent. Not loud, not hidden, just at the right level. This will generally take a fair amount of trial and error to get right but once again, depending on your context, your first thought should be to consider the merits offered by different shades of grey.
You might be starting to suspect I’m a lobbyist for the colour grey. Nobody wants to live in a world of only grey. The point is more about how its presence enables other colours to come alive. The great Bill Shankly once said ‘Football is like a piano, you need 8 men to carry it and 3 who can play the damn thing’. In data visualisation, grey does the heavy lifting so the more vibrant colours can bring the energy and vibrancy to your design.
Figure 9.25 First Fatal Accident in Spain on a High-speed Line [Translated]
Another example of the role of greyscale is demonstrated by Figure 9.25, illustrating key aspects of the tragic rail crash in Spain in 2013. The sense of foreground and background is clearly achieved by the prominence of the scarlet-coloured annotations and visual cues offset against the
471
backdrop of an otherwise greyscale palette.
Figure 9.26 Lunge Feeding
472
There are other features of annotation that will have an impact on functional harmony through their colouring. Multimedia assets like photos, embedded videos, images and illustrations need to be consistent in tone according to their relative role on the display. If they are to dominate the page then unleash the vibrancy of their colours to achieve this; if they are playing more of a secondary or supporting role then relegate their constituent colours to allow other primary features due prominence. Figure 9.26 includes small illustrations of a whale, showing how it goes through the stages of lunge feeding. The elegance of the colours used in these illustrations is entirely harmonious with the look and feel of the overall piece. They are entirely at one with the rest of the graphic. Composition: The clarity in layout of a project will often be achieved by the use of background colour to create logical organisation. In the ‘Lunge Feeding’ graphic the shading of the blue sea getting darker as it moves down is not attempting to offer a precise representation of the sea, but it gives a sense of depth and draws maximum attention to that panel. It is also naturally congruent with the subject matter.
Figure 9.27 Examples of Common Background Colour Tones
In general, there are no fixed rules on the benefits of any particular colour for background shading. Your choices will depend mostly on the circumstances and conditions in which your viewers are
473
consuming the work. Usually, when there is no associated congruence for a certain background colour, your options will tend to come from one of the selection of neutral and/or non-colours (Figure 9.27). This is because they particularly help to aid accentuation in combination with foreground colours. Typically, though, a white background (at least for your chart area) gives viewers the best chance of being able to accurately perceive the different colour attributes used in your data representation and the contrasting nature of your editorial contrast. White – or more specifically emptiness – is one of your most important options for creating functional meaning for nothingness, something I touched on earlier. The emptiness of uncoloured space can be used very effectively to direct the eye’s attention. It organises the relationship between space on a page without the need for visible apparatus, as seen in the left hand column of the lunge feeding graphic. It can also be used to represent or emphasise values that might have the state of ‘null’ or ‘zero’ to maximise contrast.
‘The single most overlooked element in visual design is emptiness. Space must look deliberately used.’ Alex White, Author, The Elements of Graphic Design
9.5 Influencing Factors and Considerations Having mapped out the ways and places where colour could be used, you will now need to consider the factors that will influence your decisions about how colour should be used.
Formulating Your Brief Format: This is a simple concern but always worth pointing out: if you are producing something for screen display you will need to set your colour output to RGB; if it is for print you will need CMYK. Additionally, when you are preparing work for print, running off plenty of proofs before finalising a design is imperative. What you are preparing digitally is a step away from the form of its intended output. What looks like a perfect colour palette on screen may not ultimately look the same when printed.
474
Print quality and consistency is also a factor. Graphics editors who create work for print newspapers or magazines will often consider using colours as close in tone as possible to pure CMYK, especially if their work is quite intricate in detail. This is because the colour plates used in printing presses will not always be 100% aligned and thus mixtures of colours may be slightly compromised. As black and white printing is still commonplace, you need to be aware of how your work might look if printed without colour. If you are creating a visualisation that might possibly be printed by certain users in black and white, the only colour property that you can feasibly utilise will be the lightness dimension. Sometimes, as a designer or author, you will be unaware of this intent and the colourful design that you worked carefully towards will end up not being remotely readable.
We all refer to black and white printing, but technically printers do not actually print using white ink, it is just less black or no black.
Furthermore, there is an important difference in how colours appear when published in colour and how they appear when published in black and white. Hues inherently possess different levels of brightness: the purest blue is darker than the purest yellow. If these were printed in black and white, blue would therefore appear a darker, more prominent shade of grey. If your printed work will need to be compatible for both colour and black and white output, before finalising your decisions check that the legibility and intended meaning of your colour choices are being maintained across both forms.
Setting: For digital displays, the conditions in which the work will be consumed will have some influence over the choice between light and dark backgrounds. The main factor is the relative contrast and the stresses this can place on the eye to adjust against the surroundings. If your work is intended for consumption in a light environment, lighter backgrounds tend to be more fitting; likewise darker backgrounds will work best for consuming in darker settings. For tablets/smartphones, the bordering colour of the devices can also influence the most suitable choice of background tone to most sympathetically contrast with the surroundings. Colour rules and identities: In some organisations there are style
475
guidelines or branding identities that require the strict use of only certain colour options. Similar guidelines may exist if you are creating work for publication in a journal, magazine, or on certain websites. Guidelines like these are well intended, driven by a desire to create conformity and consistency in style and appearance. However, in my experience, the basis of such colour guides rarely incorporates consideration for the subtleties of data visualisation. This means that the resulting palettes are often a bad fit for ideal visualisation colour needs, providing limited scope for the variation and salience you might seek to portray. Your first task should always be to find out if there is any compromise – any chance of not having these colour restrictions imposed. If there is no flexibility, then you will just have to accept this and begin acquainting yourself with the colours you do have to work with. Taking a more positive view, achieving consistency in the use of colour for visualisation within an organisation does have merits if the defined palettes offer suitably rich variety. Developing a recognisable ‘brand’ and not having to think from scratch about what colours to use every time you face a new project is something that can be very helpful, especially across a team. Purpose map: Does it need to be utilitarian or decorative? Should it be functional or appealingly seductive? Does it lend itself to being vivid and varied in colour or more muted and distinguished? Colour is the first thing we notice as viewers when looking at a visualisation, so your choices will play a huge part in setting the visible tone of voice. How you define your thinking across the vertical dimension of your purpose map will therefore have an influence on your colour thinking. Along the horizontal dimension, the main influencing consideration will be a desire to offer an ‘explanatory’ experience. As mentioned, some of the tactics for incorporating editorial salience will be of specific value if you are seeking to emphasise immediately apparent, curated insights. Ideas and inspiration: In the process of sketching out your ideas and capturing thoughts about possible sources of influence, maybe there were already certain colours you had identified as being consistent with your thinking about this subject? Additionally, you might have already identified some colours you wish to avoid using.
476
Working With Data Data examination: The characteristics of your data will naturally have a huge impact, on the decisions you make around data legibility. Firstly, the type of data you are displaying (primarily nominal vs all other types) will require a different colour treatment, as explained. Secondly, the range of categorical colour associations (limits on discernible hues) and the range and distribution of quantitative values (numbers of divisions and definition of the intervals across your classification scale) will be directly shaped by the work you did in the examination stage. In Figure 9.28 you can see a census of the prevalence and species of trees found around the boroughs of New York City. This initial big- picture view creates a beautiful tapestry made up of tree populations across the region (notice the big void where JFK Airport is located).
Figure 9.28 Excerpt from ‘NYC Street Trees by Species’
To observe patterns for individual tree types is harder: with 52 different tree species there are simply too many classifications to be able to allocate sufficiently unique colours to each. To overcome this, the project features a useful pop-up filter list which then allows you to adjust the data on view to reveal the species you wish to explore. It is often the case when thinking about colour classifications that you may need to revisit the data transformation actions to find new ways of grouping your data to create better-fit quantitative value classifications or to look at ways of grouping your categories. For the latter, actions such as combining less important categories in an
477
‘other’ bin to reduce the variability or eliminating certain values from your analysis may be necessary.
‘If using colour to identify certain data, be careful to not accidentally apply the same identity to a nearby part of the graphic. Don’t allow colour to confuse just for the sake of aesthetics. I also like to use colour to highlight. A single colour highlight on a palette of muted colours can be a strong way to draw attention to key information.’ Simon Scarr, Deputy Head of Graphics, ThomsonReuters
Establishing Your Editorial Thinking Focus: When considering the perspective of ‘focus’ in the editorial thinking stage, you were defining which, if any, elements of content would merit being emphasised. Are there features of your analysis that you might wish to accentuate? How might colour be used to accentuate key insights in the foreground and push other (less important) features into the background? What are the characteristics of your data that you might want to emphasise through changes in colour? For example, are there certain threshold values that will need to be visually amplified if exceeded? Your decisions here will directly influence your thinking about using colour to facilitate editorial salience.
Data Representation Chart type choice: Specifically in relation to data legibility, depending on which chart type you selected to portray your data, this may have attributes requiring decisions about colour. The heat map and choropleth map are just two examples that use variation in colour to encode quantitative value. Almost every chart has the potential to use colour for categorical differentiation.
Trustworthy Design Data classification: The decisions you make about how to encode data through colour have a great bearing on the legibility and accuracy of your design, especially with quantitative data. You will
478
need to ensure the classifications present a true reflection of the shape and characteristics of your data and do not suppress any significant interpretations.
‘Start with black and white, and only introduce color when it has relevant meaning. In general, use color very sparingly.’ Nigel Holmes, Explanation Graphic Designer
Meaningful: Eliminating arbitrary decisions is not just about increasing the sophistication of your design thinking, it is also an essential part of delivering a trustworthy design. If something looks visually significant in its data or editorial colouring it will be read as such, so make sure it is significant, otherwise remove it. You especially want to avoid any connotation of significant meaning across your functional or decorative colour choices. This will be confusing at best, or will appear deceptive at worst. Do not try to make something look more interesting than it fundamentally is. Colour should not be used to decorate data. You might temporarily boost the apparent appeal of your work in the eye of the viewer but this will be short-lived and artificial. Illusions: The relationship between a foreground colour and a background one can create distorting illusions that modify the perceived judgement of a colour. You saw an effect of this earlier with the inverted area chart showing ‘Gun deaths in Florida’, whereby the rising white mountain was seen by some as the foreground data, when in fact it was the background emptiness framed by the red area of data and the axis line. Illusions can affect all dimensions of colour perception. There are simply too many to mention here and they are hard to legislate for entirely; it is really more about mentioning that you need to be aware of these as a consequence of your colour choices.
Accessible Design Consistency: Consistency in the use of colours helps to avoid visual chaos and confusion and minimises cognitive effort. When you establish association through colour you need to maintain that meaning for as long as possible. Once a viewer has allocated time and effort to learn what colours represent, that association becomes
479
locked down in the eye and the mind. However, if you then allocate the same colour(s) to mean something different (within the same graphic or on a different page/screen view) this creates an additional cognitive burden. The viewer has almost to disregard the previous association and learn the new one. This demands effort that undermines the accessibility of your design.
Sometimes this can prove difficult, especially if you have a restricted colour palette. The main advice here is to try to maximise the ‘space’ between occasions of the same colour meaning different things. This space may be physical (different pages, interactive views), time (the simple duration of reading between the associations being changed) or editorial (new subject matter, new angle of analysis). Such space effectively helps to clean the palate (pun intended). Of course, at the point of any new assignment in your colour usage, clear explanations are mandatory.
Visual accessibility: Approximately 5% of the population have visual impairments that compromise their ability to discern particular colours and colour combinations. Deuteranopia is the most common form, often known as red–green colour blindness, and is a particular genetic issue associated with men. The traffic light scheme of green = ‘good’, red = ‘bad’ is a widespread approach for using colour as an indicator. It is a convenient and common metaphor and the reasons for its use are entirely understandable. However, as demonstrated in the pair of graphics in Figure 9.29, looking at some word-usage sentiment analysis, the reds and greens that most of us would easily discern (from the left graphic) are often not at all distinguishable for those with colour blindness (simulated on the right).
Figure 9.29 Demonstrating the Impact of Red-green Colour Blindness (deuteranopia)
480
Of course, if you have a particularly known, finite and fixed audience then you can easily discover if any colour-blindness issues do in fact exist. However, if your audience is much larger and undefined you are going to alienate potentially 1 person in every 14, in which case the use of the default red–green colour combination is not acceptable. Be more sensitive to your viewers by considering other options:
Figure 9.30 Colour-blind Friendly Alternatives to Green and Red
If you are working on an interactive solution, you may consider having a toggle option to switch between different colour modes. For
481
print outputs you might normally have reduced flexibility, but in certain circumstances the option of creating dual versions (second output for colour-impaired viewers) may be legitimate. Connotations and congruence: Whether it is in politics, sport, brands or in nature, there are many subjects that already have established colour associations you can possibly look to exploit. This association may sit directly with the data, such as the normal colour associations for political party categories, or more through the meaning of the data, such as perhaps through the use of green to present analysis about ecological topics. In support of accessible design, exploiting pre-existing colour associations in your work can create more immediacy in subject recognition. You might also benefit from the colour learning experiences your viewers may already have gone through. This provides a shortcut to understanding through familiarity. However, while some colour connotations can be a good thing, in some cases they can be a bad thing and possibly should be avoided. You need to be considerate of and sensitive to any colour usage to ensure that you do not employ connotations that may have a negative implication and may evoke strong emotions and reactions from people. Sometimes a colour is simply incongruent with a subject. You would not use bright, happy colours if you were portraying data about death or disease. Earlier, in the ‘Vision: Ideas’ section, I described a project context where I knew I wanted to avoid the use of blue colours in a particular project about psychotherapy treatment in the Arctic, because it would carry an unwelcome clichéd association given the subject matter. The use of ‘typical’ skin colours to represent ethnic groups in a visualisation is something that would be immediately clumsy (at best) and offensive (at worst). Cultural sensitivities and inconsistencies are also important to consider. In China, for example, red is a lucky colour and so the use of red in their stock market displays, for example, indicates the rising values. A sea of red on the FTSE or Dow Jones implies the opposite. In Western society red is often the signal for a warning or danger. Occasionally established colour associations are out of sync with contemporary culture or society. For example, when you think about colour and the matter of gender, because it has been so endlessly utilised down the years, it is almost impossible not to think instinctively about the use of blue (boys) and pink (girls). My
482
personal preference is to avoid this association entirely. I agree with so many commentators out there that the association of pink to signify the female gender, in particular, is clichéd, outdated and no longer fit for purpose. It is not too much to expect viewers to learn the association of – at most – two new colours for representing gender.
Elegant Design Unity: As I alluded to in the discussion about using colours for editorial salience, colour choices are always about contrast. The effect of using one colour is not isolated to just that instance of colour: choosing one colour will automatically create a relationship with another. There is always a minimum of two colours in any visualisation – a foreground and background colour – but generally there are many more. We notice the impact of colour decisions more when they are done badly. Inconsistent and poorly integrated colour combinations create jarring and discordant results. If we do not consciously notice colour decisions this probably means they have been seamlessly blended into the fabric of the overall communication. Neutral colouring: Even if there is no relevance in the use of colour for quantitative or categorical classifications, you still have to give your chart some colour, otherwise it will be invisible. The decision you make will depend again on the relative harmony with other colour features but should also avoid unnecessarily ‘using up’ a useful colour. Suppose you colour your bars in blue but then elsewhere across your visualisation project blue would have been a useful colour to show something meaningful; you then have unnecessarily taken blue out of the reckoning. My default choice is to go with grey to begin with (Figure 9.31) and only use a colour if there is a suitable and available colour not used elsewhere or if it needs to be left as a back- or mid-ground artefact to preserve prominence elsewhere in the display.
Figure 9.31 Excerpt from ‘Pyschotherapy in The Arctic’
483
Justified: Achieving elegant design is about eliminating the arbitrary. In thinking about colour usage I often get quite tough with myself. If I want to show any feature on my visualisation display I have to seek permission from myself to unlock access to the more vibrant colours by justifying why I should be allowed to use and apply that colour (I know what you’re thinking, ‘what a fun existence this guy leads’). Elegance in visualisation design is often about using only the colours you need to use and avoiding the temptation to inject unnecessary decoration. The Wind Map project (Figure 9.32) demonstrates unquestionable elegance and yet uses only a monochromatic palette. There is no colouring of the sea, no topographic detail, no emphasising of any extreme wind speed thresholds being reached. The resulting elegance is quite evident: the map has artistic and functional beauty. To emphasise again, I am not advocating a need to pursue minimalism: while you can create incredibly elegant and detailed works from a limited palette of colours, justifying the use of colours is not the same as unnecessarily restricting the use of colour. Feels right: The last component of influence is yourself. Sometimes you will just find colours that feel right and look good when you apply them to your work. There is maybe no underlying science behind such choices, and as such you will simply need to back your own instinctive judgement as an astute visualiser and know when something looks good. Creating the right type of visual appeal, something that is pleasing to the eye and equally fit for purpose in all
484
the functional ways I have outlined, is a hard balance to achieve, but you will find that weighing up all these different components of influence alongside your own flair for design judgement will give you the best chance of getting there.
Figure 9.32 Wind Map
Summary: Colour Data legibility involves using colours to represent different types of data. The most appropriate colour association or scale decisions will depend on the data type: nominal (qualitative), ordinal (qualitative), interval and ratio (quantitative).
Editorial salience is about using colour to direct the eye. For which features and to what degree of emphasis do you want to create contrast?
Functional harmony concerns deciding about every other colour property as applied to all interactive features, annotations and aspects of your composition thinking.
Influencing Factors and Considerations
Formulating the brief: format, setting, colour rules and imposed guidelines all have a significant impact. Your definitions about both tone and and experience, on the purpose map, will lead to specific
485
choices being more suitable than others. What initial ideas did you form? Have any sources of inspiration already implanted ideas inside your head about which colours you could use? Working with data: what type of data and what range of values/number of classifications have you got? Establishing your editorial thinking: what things do you want to emphasise or direct the eye towards (focus)? Data representation: certain chart type choices will already include colour as an encoded attribute. Trustworthy design: ensure that your colour choices are faithful to the shape of your data and the integrity of your insights. If something looks meaningful it should be, otherwise it will confuse or deceive. Accessible design: once you’ve committed colour to mean something preserve the consistency of association for as long as possible. Be aware of the sensitivities around visual accessibility and positive/negative colour connotations. Elegant design: the perception of colours is relative so the unity of your choices needs to be upheld. Ensure that you can justify every dot of colour used and, ultimately, rely on your own judgment to determine when your final palette feels right.
Tips and Tactics
Use the squint test: shrink things down and/or half close your eyes to see what coloured properties are most prominent and visible – are these the right ones? Experimentation: trial and error is still often required in colour, despite the common sense and foundation of science attached to it. Developing a personal style guide for colour usage saves you the pain of having to think from scratch every time and will help your work become more immediately identifiable (which may or may not be an important factor). Make life easier by ensuring your preferred (or imposed) colour palettes are loaded up into any tool you are using, even if it is just the tool you are using for analysis rather than for the final presentation of your work. If you are creating for print, make sure you do test print runs of the draft work to see how your colours are looking – do not wait for the first print when you (think you) have finished your process.
486
487
10 Composition
Composition concerns making careful decisions about the physical attributes of, and relationships between, every visual property to ensure the optimum readability and meaning of the overall, cohesive project.
Composition is the final layer of your design anatomy, but this should not imply that it is the least important part of your design workflow. Far from it. It is simply that now is the most logical time to think about this, because only at this point will you have established clarity about what content to include in your work. As I explained, this final layer of design thinking, along with colour, is no longer about what elements will be included but how they will appear. Composition is a critical component of any design discipline. The care and attention afforded in the precision of your composition thinking will continue until the final dot or pixel has been considered.
Visual assets such as your chart(s), interactive controls and annotations all occupy space. In this chapter you will be judging what is the best way to use space in terms of the position, size and shape of every visible property. In many respects these individual dimensions of thought are inseparable and so, similar to the discussion about annotation, the division in thinking is separated between project- and chart-level composition options:
Project composition: defining the layout and hierarchy of the entire visualisation project. Chart composition: defining the shape, size and layout choices for all components within your charts.
10.1 Features of Composition: Project Composition This first aspect of composition design concerns how you might lay out and size all the visual content in your project to establish a meaningful hierarchy and sequence. Content, in this case, means all of your charts, interactive operations and elements of annotation.
488
Where will you put all of this, what size will it be and why? How will the hierarchy (across views) and sequencing (within a view) best fit the space you have to work in? How will you convey the relative importance and provide a connected narrative where necessary?
I will shortly run through all the key factors that will influence your decisions, but it is worth emphasising that so much about composition thinking is rooted in common sense and involves a process of iteration towards what feels like an optimum layout. Of course, there are certain established conventions, such as the positioning of titles first or at the top (usually left or centrally aligned). Introductions are inevitably useful to offer early, whereas footnotes detailing data sources and credits might be of least importance, relatively speaking. You might choose to show the main features first, exploiting the initial attention afforded by your audience, or you may wish to build up to this, starting off with contextual content before the big ‘reveal’.
Figure 10.1 City of Anarchy
489
490
The hierarchy of content is not just a function of relative position through layout design, it can also be achieved through the relative variation in size of the contents. Just as variation in colour implies significance, so too does variation in size: a chart that is larger than another chart will imply that the analysis it is displaying carries greater importance.
The ‘City of anarchy’ infographic demonstrates a clear visual hierarchy across its design. There is a primary focal point of the main subject ‘cutaway’ illustration in the centre with a small thumbnail image above it for orientation. At the bottom there are small supplementary illustrations to provide further information. It is clear through their relative placement at the bottom of the page and their more diminutive stature that they are of somewhat incidental import compared with the main detail in the centre.
There are generally two approaches for shaping your ideas about this project-level composition activity, depending on your entry-point perspective: wireframing and storyboarding. I profiled these at the start of this part of the book, but it is worth reinforcing their role now you are focusing on this section of design thinking.
Wireframing involves sketching the potential layout and size of all the major contents of your design thinking across a single-page view. This might be the approach you take when working on an infographic or any digital project where all the interactive functions are contained within a single-screen view rather than navigating users elsewhere. Any interactive controls included would have a description within the wireframe sketch to explain the functions they would trigger. Figure 10.2 is an early wireframe drawn by Giorgia Lupi when shaping up her early thoughts about the potential layout of a graphic exploring various characteristics of Nobel prizes and laureates between 1901 and 2012.
Figure 10.2 Wireframe Sketch
491
Storyboarding is something you would undertake with wireframing if you have a project that will entail multiple pages or many different views and you want to establish a high-level feel for the overall architecture of content, its navigation and sequencing. This would be an approach relevant for linear outputs like discrete sequences in reports, presentation slides or video graphics, or for non-linear navigation around different pages of a multi-faceted interactive. The individual page views included as cells in this big-picture hierarchy will each merit more detailed wireframing versions to determine how their within-page content will be sized and arranged, and how the navigation between views would operate. With both wireframing and storyboarding activities all you are working towards, at this stage, are low-fidelity sketched concepts. Whether this sketching is on paper or using a quick layout tool does not matter; it just needs to capture with moderate precision the essence of your early thinking about the spatial consequence of bringing all your design choices together. Gradually, through further iteration, the precision and finality of your solution will emerge.
492
10.2 Features of Composition: Chart Composition After establishing your thoughts about the overall layout, you will now need to go deeper in your composition thinking and contemplate the detailed spatial matters local to each chart, to optimise its legibility and meaning. There are many different components to consider.
Chart size: Do not be afraid to shrink your charts. The eye can still detect at quite small resolution and with great efficiency chart attributes such as variation in size, position, colour, shape and pattern. This supports the potential value of the small-multiples technique, an approach that tends to be universally loved in data visualisation. As I explained earlier, this technique offers an ideal solution for when you are trying to display the same analysis for multiple categories or multiple points in time. Providing all the information in a simultaneous view means that viewers can efficiently observe overall patterns as well as perform a more detailed inspection. Figure 10.3 provides a single view of a rugby team’s match patterns across the first 12 matches of a season. Each line chart panel portrays the cumulative scoring for the competing teams across the 80 minutes of a match. The 12 match panels are arranged in chronological order, from top left to bottom right, based on the date of the match.
Figure 10.3 Example of the Small Multiples Technique
493
The main obstacle to shrinking chart displays is the impact on text. The eye will not cope too well with small fonts for value or category labels, so there has to be a trade-off, as always, between the amount of detail you show and the size you show it. Chart scales: When considering your chart-scales try to think about how you might use these to tell the viewer something meaningful. This can be achieved through astute choices around the maximum value ranges and also in the choice of suitable intervals for labelling and gridline guides. The maximum values that you assign to your chart scales, informed by decisions around editorial framing, can be quite impactful in surfacing key insights. You may recall the chart from earlier that looked at the disproportionality of women CEO’s amongst the S&P 1500 companies. Figure 10.4 is another graphic on a similar subject, which contextualises the relative progress in the rise of women CEOs amongst the Fortune 500 companies. By setting the maximum y-axis value range to reflect the level at which equality would exist, the resulting empty space emphasises the significant gap that still persists.
Figure 10.4 Reworking of ‘The Glass Ceiling Persists’
494
Figure 10.5 Fast-food Purchasers Report More Demands on Their Time
495
Figure 10.5 shows how the lack of careful thought about your scales can undermine the ease of readability. This chart shows how American adults spend their time on different activities. The analysis is broken down into minutes and so the maximum is set at 1440 minutes in a day. For some reason, the y-axis labels and the associated horizontal gridlines are displayed at intervals of 160 minutes. This is an entirely meaningless quantity of time so why divide the day up into nine intervals? To help viewers perceive the significance and size of the different stacked activities it would have been far more logical to use 60-minute time intervals as that is how we tend to think when dividing our daily schedule. Chart orientation: Decisions about the orientation of your chart and its contents can sometimes help squeeze out an extra degree of readability and meaning from your display.
Figure 10.6 Illustrating the Effect of Chartorientation Decisions
496
The primary concern about chart orientation is towards the legibility of labels along the axis. A vertical bar chart, with multiple categories along the x-axis, will present a challenge of making the labels legible and avoiding them overlapping. Ideally you would want to preserve label reading in line with the eye, but you might need to adjust their orientation to either 45° or 90°. My preference for handling this with bar charts is to switch the orientation of the chart and to then have
497
much more compatible horizontal space to accommodate the labels. The meaning of your subject’s data may also influence your choice. While there may have been constraints on the dimension of space in its native setting, Figure 10.6, portraying the split of political parties in Germany, feels like a missed opportunity to display a political axis of the Left and the Right through using a landscape rather than portrait layout. As you saw earlier, the graphic about ‘Iraq’s bloody toll’ (Figure 1.11) uses an inverted bar chart to create a potent display of data that effectively conveys the subject matter, but importantly does so without introducing any unnecessary obstacles in readability. In the previous section I presented a wireframe sketch of a graphic about Nobel prize winners. Figure 10.7 shows the final design. Notice how the original concept of the novel diagonal orientation was accomplished in the final composition, exploiting the greater room that this dimension of space offers within the page. It feels quite audacious to do this in a newspaper setting.
Figure 10.7 Nobels no Degrees
498
Figure 10.8 Kasich Could Be The GOP’s Moderate Backstop
Figure 10.8, from FiveThirtyEight, rotates the scatter plot by 45° and then overlays a 2 × 2 grid which helps to guide the viewer’s interpretation by making it easier to observe which values are located in each quadrant. It is also used to emphasise the distinction between location in the top and bottom halves of the chart along the axis of popularity, essentially the primary focus of the analysis.
Although the LATCH and CHRTS acronyms share some similarities, the application of each concerns entirely different aspects of your design thinking. They are independent of one another. A bar chart, which belongs to the categorical (C) family of charts, could have its data potentially sorted by location, alphabet, time, category or hierarchy.
499
Chart value sorting: Sorting content within a chart is important for helping viewers to find and compare quickly the most relevant content. One of the best ways to consider the options for value sorting comes from using the LATCH acronym, devised by Richard Saul Wurman, which stands for the five ways of organising displays of data: Location, Alphabet, Time, Category or Hierarchy. Location sorting involves sequencing content according to the order of a spatial dimension. This does not refer to sorting data on a map locations are fixed, rather it could be sorting data by geographical spatial relationships (such as presenting data for all the stops along a subway route) or a non-geographical spatial relationship (like a sequence based on the position of major parts of the body from head to toe). You should order by location only when you believe it offers the most logical sequence for the readability of the display or if there is likely to be interest or significance in the comparison of neighbouring values. An example of location sorting is displayed in ‘On Broadway’ (Figure 10.9) on the following page, an interactive installation that stitches together a sequenced compilation of data and media related to 30 metre intervals of life along the 13 miles (21 km) of Broadway that stretches across the length of Manhattan. This continuous narrative offers compelling views of the fluctuating characteristics as you transport yourself down the spine of the city.
Figure 10.9 On Broadway
Alphabetical sorting is a cataloguing approach that facilitates efficient lookup and reference. Only on rare occasions, when you are
500
especially keen to offer convenient ordering for looking up categorical values, will you find that alphabetical sorting alone offers the best sequence. In Figure 10.10, investigating different measures of waiting times in emergency rooms across the United States, the bar charts are presented based on the alphabetical sorting of each state. This is the default setting but users can also choose to reorder the table hierarchically based on the increasing/decreasing values across the four columns.
Data representation techniques that display overlapping connections, like Sankey diagrams, slope graphs and chord diagrams, also introduce the need to contemplate value sorting in the z-dimension: that is, which of these connections will be above and which will be below, and why.
Alphabetical sorting might be seen as a suitably diplomatic option should you not wish to imply any ranking significance that would be displayed when sorting by any other dimension. Additionally, there is a lot of sense in employing alphabetical ordering for values listed in dropdown menus as this offers the most immediate way for viewers to quickly find the options they are interested in selecting.
Figure 10.10 ER Wait Watcher: Which Emergency Room Will See You the Fastest?
501
Time-based sorting is used when the data has a relevant chronological sequence and you wish to display and compare how changes have progressed over time. In Figure 10.11, you can see a snapshot of a graphic that portrays the rain patterns in Hong Kong since 1990. Each row of data represents a full year of 365/366 daily readings running from left to right. The subject matter and likely interest in the seasonality of patterns make chronological ordering a common-sense choice.
Figure 10.11 Rain Patterns
Categorical sorting can be usefully applied to a sequence of categories that have a logical hierarchy implied by their values or unique to the subject matter. For example, if you were presenting analysis about football players you might organise a chart based on the general order of their typical positions in a team (goalkeeper > defenders > midfielders > forwards) or use seniority levels as a way to present analysis about staff numbers. Alternatively, if you have ordinal data you can logically sort the values according to their inherent hierarchy. In Figure 10.12, that you saw earlier in the profile of ordinal colours, the columns are sequenced left to right in order from ‘major deterioration’ to ‘major improvement’, to help reveal the balance of treatment outcomes from a sample of psychotherapy clients.
Figure 10.12 Excerpt from ‘Pyschotherapy in The Arctic’
502
Hierarchical sorting organises data by increasing or decreasing quantities so a viewer can efficiently perceive the size, distribution and underlying ranking of values. In Figure 10.13, showing the highest typical salaries for women in the US, based on analysis of data from the US Bureau of Labour Statistics, the sorting arrangement presents the values by descending quantity to reveal the highest rankings values.
Figure 10.13 Excerpt from ‘Gender Pay Gap US’
In Figure 10.12 the bubbles in each column do not need to be coloured as their position already provides a visual association with the ‘deterioration’ through to ‘improvement’ ordinal categories. The attribute of colour, specifically, can therefore be considered redundant encoding. However, you might still choose to include this redundancy if
503
you believed it aided the immediacy of association and distinction. In this case, the chart was part of a larger graphic that employed the same colour associations across several different charts and therefore it made sense to preserve this association.
10.3 Influencing Factors and Considerations You are now familiar with the array of various aspects of composition thinking. At this point you will need to weigh up your decisions on how you might employ these in your own work. Here are some of the specific factors to bear in mind.
Formulating Your Brief Format: Naturally, as composition is about spatial arrangement, the nature and dimensions of the canvas you have to work with will have a fundamental bearing on the decisions you make. There are two concerns here: what will be the shape and size of the primary format and how transferable will your solution be across the different platforms on which it might be used or consumed? Another factor surrounding format concerns the mobility of viewing the work. If the form of your output enables viewers to easily move a display or move around a display in a circular plane (such as looking at a printout or work on a tablet) this means that issues such as label orientation can be largely cast aside. If your output is going to be consumed in a relatively fixed setting (desktop/laptop or via a presentation) the flexibility of viewing positions will be restricted.
Working With Data Data examination: Not surprisingly, the shape and size of your data will directly influence your chart composition decisions. When discussing physical properties in Chapter 4, I described the influence of quantitative values with legitimate outliers distorting ideal scale choices. One solution for dealing with this is to use a non-linear logarithmic (often just known as a ‘log’) scale. Essentially, each major interval along a log scale increases the value at that marked position by a factor of 10 (or by one order of magnitude) rather than
504
by equal increments. In Figure 10.14, looking at ratings for thousands of different board games, the x-axis is presented on a log scale in order to accommodate the wide range of values for the ‘Number of ratings’ measure and to help fit the analysis into a square-chart layout. Had the x-axis remained as a linear scale, to preserve a square layout would have meant squashing values below 1000 into such a tightly packed space that you would hardly see the patterns. Alternatively, a wide rectangular chart would have been necessary but impractical given the limitations of the space this chart would occupy. I have great sympathy for the challenges faced by designers like Zimbabwe-based Graham van de Ruit, when working on typesetting a book titled Millions, Billions, Trillions: Letters from Zimbabwe, 2005−2009 in 2014. The book was all text, apart from one or two tables. One of the tables of data supplied to Graham showed Zimbabwe’s historical monthly inflation rates, which, as you can see (Figure 10.15), included some incredibly diverse values. I love the subtle audacity of Graham’s solution. Even though it is presented in tabular form there is a strong visual impact created by allowing the sheer spatial consequence of the exceptional mid-2008 numbers to cause the awkward widening of the final column. I think this makes the point much more effectively than a chart might, in this case.
Figure 10.14 The Worst Board Games Ever Invented
505
Figure 10.15 From Millions, Billions, Trillions: Letters from Zimbabwe, 2005−2009
506
‘I thought that a graph might be more effective, but I quickly realised that the scale would be a big challenge… The whole point of graphing would have been to show the huge leap in 2008, something that I felt the log scale would detract from and was impractical with the space constraints. I also felt that a log scale might not be intuitive to the target audience.’ Graham van de Ruit, Editorial and Information Designer
Establishing Your Editorial Thinking Angles: The greater the number of different angles of analysis you wish to cover in your work, the greater the challenge will be to seamlessly accommodate the resulting chart displays in one view. The more content you include increases the need to contemplate reductions in the size of charts or a non-simultaneous arrangement, perhaps through multi-page sequences with interactive navigation. In defining your editorial perspectives, you will have likely established some sense of hierarchy that might inform which angles should be more prominent (regarding layout position and size) and which less so. There might also be some inherent narrative binding
507
each slice of analysis that lends itself to being presented in a deliberate sequence.
Data Representation Chart type choice: Different charts have different spatial consequences. A treemap generally occupies far more space than a pie chart simply because there are many more ‘parts’ being shown. A polar chart is circular in shape, whereas a waffle chart is squared. With each chart you include you will have a uniquely shaped piece that will form part of the overall jigsaw puzzle. Inevitably there will be some shuffling of content to find the right size and placement balance. The table in Figure 10.16 summarises the main chart structures and the typical shapes they occupy. This list is based only on the charts included in the Chapter 6 gallery but still offers a reasonable compilation of the main structures. These are ordered in descending frequency as per the distribution of the different structures of charts in the gallery.
Figure 10.16 List of chart structures
508
Trustworthy Design Chart-scale optimisation: Decisions about chart scales concern the maximum, minimum and interval choices that ensure integrity through the representation as well as optimise readability. Firstly, let’s look at decisions around minimum values used on the quantitative value axis, known as the origin, and the reasons why it is not OK for you to truncate the axis in methods like the bar chart. Any data representation where the attribute of size is used to encode a quantitative value needs to show the full, true size, nothing more and nothing less. The origin needs to be zero. When you truncate a bar chart’s quantitative value axis you distort the perceived length or height of the bar. Visualisers are often tempted to crop axis scales when values are large and the differences between categories are small. However, as you can see in Figure 10.17, the consequence is that it creates the impression of highly noticeable relative difference between values when the absolute values do not support this.
509
Figure 10.17 Illustrating the Effect of Truncated Bar Axis Scales
The single instance in which it is remotely reasonable to truncate an axis would be if you had a main graphic which effectively offered a thumbnail view of the whole chart for orientation positioned alongside a separate associated chart (similar to that on the right). This separate chart might have a truncated axis that would provide a magnified view of the main chart, showing just the tips of the bar, to help viewers see the differences close up. In contrast to the bar chart, a line chart does not necessarily need always to have a zero origin for the value axis (normally the y-axis). A line chart’s encoding involves a series of connected lines (marks) joining up continuous values based on their absolute position along a scale (attribute). It therefore does not encode quantitative values through size, like the bar chart does, so the truncation of a value axis will not unduly impact on perceiving the relative values against the scale and the general trajectory. For some data contexts the notion of a zero quantity might be impossible to achieve. In Figure 10.18, showing 100m sprint record times, no human is ever going to be able to run 100m in anywhere near zero seconds. Times have improved, of course, but there is a physical limit to what can be achieved. To show this analysis with the y-axis starting from zero would be unnecessary and even more so if you plotted similar analysis for longer distance races. However, if you were to plot the 100m results and the 400m results on the same chart, you would need to start from zero to enable orientation of the scale of comparable values. This sense of comparable scale is missing from the next chart, whereby including the full quantitative value range down to zero would be necessary to perceive the relative scale of attitudes towards same-sex marriage. The chart’s y-axis appears to start from an origin of 20 but as we are
510
looking at part-to-whole analysis, the y-axis should really be displayed from an origin of zero. The maximum doesn’t need to go up to 100%, the highest observed value is fine in this case, but it could be interesting to set the maximum range to 100% in order to create a similar sense of the gap to be bridged before 100% of respondents are in agreement.
Figure 10.18 Excerpt from ‘Doping under the Microscope’
Figure 10.19 Record-high 60% of Americans Support Same-sex Marriage
Aspect ratios: The aspect ratio of a line chart, as derived from the height and width dimensions of the chart area, can have a large impact on the perceived trends presented. If the chart is too narrow,
511
the steepness of connections will be embellished and look more significant; if the chart is stretched out too wide, the steepness of slopes will be much more dampened and key trends may be somewhat disguised. There is no absolutely right or wrong approach here but clearly there is a need for sensitivity to avoid the possibility of unintended deception. A general rule of thumb is to seek a chart area that enables the average slope to be presented at 45°, though this is not something that can be easily and practically applied, especially as there are many other variables at play, such as the range of quantitative and time values and the scales being used. My advice is just to make a pragmatic judgement by eye to find the ratio that you think is faithful to the significance of the trends in your data. Mapping projections: One of the most contentious matters in the visual representation of data relates to thematic mapping and specifically to the choice of map projection used. The Earth is not flat (hopefully no contention there, otherwise this discussion is rather academic), yet the dominant form through which maps are presented portrays the Earth as being just that. Features such as size, shape and distance can be measured accurately on Earth but when projected on a flat surface a compromise has to occur. Only some of these qualities can be preserved and represented accurately.
I qualify this with ‘dominant’ because, increasingly, advances in technology (such as WebGL) mean we can now interact with spherical portrayals of the Earth within a 2D space.
There are lots of exceptionally complicated calculations attached to the variety of spatial projections. The main things you need to know about projection mapping are that:
every type of map projection has some sort of distortion; the larger the area of the Earth portrayed as a flat map, the greater the distortion; there is no single right answer – it is often about choosing the least-worst case.
Thematic mapping (as opposed to mapping spatially for navigation or reference purposes) is generally best portrayed using mapping projections based on ‘equal-area’ calculations (so the sacrifice is more on the shape, not the size). This ensures that the phenomena per unit – the values you are
512
typically plotting – are correctly represented by proportion of regional area. For choosing the best specific projection, in the absence of perfect, damage limitation is often the key: that is, which choice will distort the spatial truth the least given the level of mapping required. There are so many variables at play, however, based on the scope of view (world, continent, or country/sub-region), the potential distance from the equator of your region of focus and whether you are focusing on land, sea or sky (atmosphere), to name but a few. As with many other topics in this field, a discussion about mapping projections requires a dedicated text but let me at least offer a brief outline of five different projections to begin your acquaintance:
Many tools that offer rudimentary mapping options will tend to only come with a default (non-adjustable) projection, often the Mercator (or Web Mercator). The more advanced geospatial analysis tools will offer pre-loaded or add-in options to broaden and customise the range of projections. Hopefully, in time, an increasing range of the more pragmatic desktop tools will enhance projection customisations.
Figure 10.20 A Selection of Commonly Deployed Mapping Projections
513
Accessible Design Good design is unobtrusive: One of the main obstructions to facilitating understanding through a visualisation design is when viewers are required to rely on their memory to perform comparisons between non-simultaneous views. When the composition layout requires viewers to flick between pages or interactively generated views, they have to try store one view in their mind and then mentally compare that against the live view that
514
has arrived on the screen. This is too hard and too likely to fail given the relatively weak performance of the brain’s working memory. Content that warrants direct comparison should be enabled through proximity to and alignment with related items. I mentioned in the section on animation that if you want to compare different states over time, rather than see the connected system of change, you will need to have access to the ‘moment’ views simultaneously and without a reliance on memory.
‘Using our eyes to switch between different views that are visible simultaneously has much lower cognitive load than consulting our memory to compare a current view with what was seen before.’ Tamara Munzner taken from Visualization Analysis and Design
Elegant Design
‘I’m obsessed with alignments. Sloppy label placement on final files causes my confidence in the designer to flag. What other details haven’t been given full attention? Has the data been handled sloppily as well? … On the flip side, clean, layered and logically built final files are a thing of beauty and my confidence in the designer, and their attention to detail, soars.’ Jen Christiansen, Graphics Editor at Scientific American
Unity: As I discussed with colour, composition decisions are always relative: an object’s place and its space occupied within a display immediately create a relationship with everything else in the display. Unity in composition provides a similar sense of harmony and balance between all objects on show as was sought with colour. The flow of content should feel logical and meaningful. The enduring idea that elegance in design is most appreciated when it is absent is just as relevant with composition. Look around and open your eyes to composition that works and does not work, and recognise the solutions that felt effortless as you read them and those that felt punctured and confusing. This is again quite an elusive concept and one that only comes with a mixture of common-sense judgement, experience and exposure to inspiration from elsewhere. Thoroughness: Precision positioning is the demonstration of
515
thoroughness and care that is so important in the pursuit of elegance. You should aim to achieve pixel-perfect accuracy in the position and size of every single property. Think of the importance of absolute positioning in the context of detailed architectural plans that outline the position of every fine detail down to power sockets, door handles and the arc of a window’s opening manoeuvre. A data visualiser has to commit to ultimate precision and consistency because any shortcomings will be immediately noticeable and will fundamentally impact on the function of the work. If you do not feel a warm glow from every emphatic snap-to-grid resize operation or upon seeing the results of a mass alignment of page objects, you are not doing it right. (Honestly, I am loads of fun to be around.
Summary: Composition Project composition defines the layout and hierarchy of the entire visualisation project and may include the following features:
Visual hierarchy – layout: how to arrange the position of elements? Visual hierarchy – size: how to manage the hierarchy of element sizes? Absolute positioning: where specifically should certain elements be placed?
Chart composition defines the shape, size and layout choices for all components within your charts and may include the following features:
Chart size: don’t be afraid to shrink charts, so long as any labels are still readable, and especially embrace the power of small multiple. Chart scales: what are the most meaningful range of values given the nature of the data? Chart orientation: which way is best? Chart value sorting: consider the most meaningful sorting arrangement for your data and editorial focus, based on the LATCH acronym.
Influencing Factors and Considerations
Formulating the brief: what space have you got to work within?
516
Working with data: what is the shape and size of your data and how might this affect your chart design architecture? Establishing your editorial thinking: how many different angles (charts) might you need to include? Is there any specific focus for these angles that might influence a sequence or hierarchy between them? Data representation: any chart has a spatial consequence – different charts have different structures that will create different dimensions that will need to be accommodated. Trustworthy design: the integrity and meaning of your chart scale, chart dimensions, and (for mapping) your projection choices are paramount. Accessible design: remember that good design is unobtrusive – if you want to facilitate comparisons between different chart displays these ideally need to be presented within a simultaneous view. Elegant design: unity of arrangement is another of the finger-tip sense judgments but will be something achieved by careful thinking about the relationships between all components of your work.
Tips and Tactics
You will find that as you reach the latter stages of your design process, the task of nudging things by fractions of a pixel and realigning features will dominate your attention. As energy and attention start to diminish you will need to maintain a commitment to thoroughness and a pride in precision right through to the end! Empty space is like punctuation in visual language: use it to break up content when it needs that momentary pause, just as how a comma or full stop is needed in a sentence. Do not be afraid to use empty space more extensively across larger regions as a device to create impact. Like the notes not played in jazz, effective visualisation design can also be about the relationship between something and nothing.
517
518
- Part C Developing Your Design Solution
- 6 Data Representation
- 7 Interactivity
- 8 Annotation
- 9 Colour
- 10 Composition
