While there is an art to dashboard design, the science behind visualization is more important than most people realize. In these webinar slides, we show how to approach data visualization in a systematic way that will unlock the story your data holds.
2. How do you create
dashboards that users
will adopt and use?
Is there a way to
determine effectiveness of
data visualizations?
How do I
communicate the
story in my data?
Why is data
visualization
important?
3. WHY IS DATA VISUALIZATION IMPORTANT?
WHY IS DATA VISUALIZATION IMPORTANT?
4. WHY IS DATA VISUALIZATION IMPORTANT?
COGNITIVE LIFT
56. Decrease the cognitive lift.
Think twice before I visualize information.
What is my purpose? Exploratory or explanatory?
If explanatory, what is my one-sentence summary of my sto
Aesthetics matter.
Mange the tension.
57. References & Recommended Resources
“Storytelling with Data” by Cole Nussbaumer Knaflic
“Signal” by Stephen Few
“Now You See It” by Stephen Few
“Data Points” by Nathan Yau
d3js.org Visualization Examples
“Calendar View” chart https://bl.ocks.org/mbostock/4063318 via d3js.org
“Time Series” chart http://square.github.io/cubism/ via d3js.org
“Gantt” chart http://bl.ocks.org/dk8996/5538271 via d3js.org
“Bullet” chart http://bl.ocks.org/CodeXmonk/6112167 via d3js.org
“Force-Directed” chart https://bl.ocks.org/mbostock/4062045 via d3js.org
“Dependency Wheel” chart http://www.redotheweb.com/DependencyWheel/ via d3js.org
Cited and leveraged for presentation
58. Decrease the cognitive lift.
Think twice before I visualize information.
What is my purpose? Exploratory or explanatory?
If explanatory, what is my one-sentence summary of my sto
Aesthetics matter.
Mange the tension.
Questions?
Whether you are a business leader, a strategist in your organization, an analyst, or a developer – data matters to you now more than ever. My goal today is to convince you that how you visualize your data is key to unlocking and unleashing its full potentialWe will discuss the answers to these questions today. Please note that all resources I reference will be provided to you to further your learning after this webinar.
We begin with a simple question. Why is data visualization important?One answer is surprisingly simple – humans are really good at understanding data when visualized properly. We can process large amounts of information and determine what is most important. We can categorize and group, identify trends, and eliminate noise with our brains without much effort when we use our visual system to interpret data.
When used correctly, data visualization can reduce the time it takes to process information.
Because of this, it is our responsibility as data visualizers to minimize the work required of our audience – Ask “how hard am I making people work to understand this information?” Another word for this is cognitive lift.
Anyone visualizing data should obsess over how to minimize this burden on their audience at all costs. If there’s one thing I want you to take away from this webinar, its that this goal should be your main purpose when visualizing data. Obsess over how to minimize cognitive lift for your audience at all costs.In her book, “Storytelling with Data”, Cole Nussbaumer Knaflic (Kole Nuse-Bow-Mer Naff-Lic) argues that “every single element that [we] put on a page takes up cognitive load on the part of [our] audience” (pg. 14). This lift requires time and effort, especially when it comes to understanding data.
There are systematic and objective ways we can minimize the stress we put on our audience by using what are called pre-attentive attributes, which are what our brain does behind the scenes when we first see a visual - she focuses on six attributes, which we will cover now.
I’ll give you a moment to interpret these visuals.
Did you notice how the same data is repeated five times? Even still, your brain interpreted the data five different ways without much thought.
This is an example of how our pre-attentive attributes work - we can use these powerful tools to bring attention to what we think is most important in our data. Our brains not only interpret relationships, but it prioritizes them as well. Did you notice how you interpreted the connectedness of the dots as a higher priority attribute than their proximity to one another?
Effective understanding and use of these attributes is key, as unintended relationships result when they are used inappropriately. You don’t want to communicate relationships that don’t exist – you incur a cost in cognitive load with your audience to undo inferred relationships.
Stephen Few, in his book “Now You See It” goes a step further and describes how our pre-attentive attributes can help interpret not only relationships in data, but also quantitative values of the data a visualization represents - though these representations carry varying degrees of precision. For example, we intuitively interpret longer lines as being “greater” than shorter lines, and dot position up and to the right as being “greater” than those down and to the left.
He argues that our brains are very precise at interpreting these two attributes, and are not very precise when interpreting the other 4 - how wide, how large, how intense, and how blurry objects are.
Greater precision decreases cognitive lift. It is easier to interpret how much longer something is than how much wider something is – we need to keep this in mind with data visualization. Make it easier for your audience at all costs.
Two final pre-attentive attributes here from “Storytelling with Data” that can minimize cognitive lift.
Closure and continuity emphasize how our brain “completes the picture”. We experience a complete circle and a Y-Axis on this bar chart, though neither exist visually. We can use these attributes for our main purpose stated earlier - to eliminate clutter and reduce cognitive load in our visualizations. No need to include them if our audience’s brains will provide them for us.
Here’s a tip:
Before presenting data to your audience, ask yourself “How can I decrease cognitive lift required by my audience?”
Another way of putting that – are pre-attentive attributes working for me or against me on my visualizations? Can I “take out the trash” in my presentation and utilize pre-attentive attributes of my audience to eliminate clutter?
Speaking of reducing cognitive load… dulling this background picture helped decrease the work your brain had to do to focus on the text!
A different answer to the question of Why data is visualization important? is equally relevant though.
Data is a numeric or textual representation of the facts. Because data visualizations are one level removed from data – as representations of data themselves, visualizations can remind people (albeit sub-consciously) that the data itself is just a representation of people, places and things. We could visualize data in many different ways, and that may lead us to different conclusions. Similarly, new data sets will provide new contexts that may lead us to different conclusions as well.
Understanding that our data has a story behind it should help us figure out the best way to craft a visualization.
Like any good story, context matters. Nathan Yau in his book “Data Points” says – that “without context, data is useless, and any visuals you create with it will be useless”
Knowing the story behind your data will help you determine how to explain the story in your data.
Yau encourages us in “Data Points” to ask who collected the data, how they collected it, and why they collected it, as this will tell us the reliability of the data, potential biases in the data, and expose any ethical concerns in the data. Ask yourself, “based on the nature of my data, what can I reliably put emphasis on? What are valid conclusions, and what are invalid conclusions someone could draw from this data? If a visual has the risk of misleading someone, exposing someone, or purposefully harming someone – how should I present the information?”
Think twice how you visualize and communicate that information.
An example he uses to demonstrate this principle is visualizing data from the 2010 Gawker password breach . He says “the mean thing to do would be to highlight usernames with common/poor passwords, or… create an application that guessed passwords”.
In the end, he proposes visualizing the common passwords from the breach to help serve as a warning to users to make their passwords less obvious, and provides context for why terms appeared to draw conclusions. People tended to use the name of the Gawker sub-sites like lifehack or Gizmodo as their passwords – this visualization helps draw attention to that fact to cause people to think twice before using these passwords.
I like the quote he provides that “Data is an abstraction of real life, and real life can be complicated, but if you gather enough context, you can at least put forth a solid effort to make sense of it.”
Data Visualization and communication can have positive or negative effects, but either way the effect can be powerful because of the information it conveys. That is why it is important how we approach visualization.Also, stop using “password” as your password, folks.
This leads us to our next question – how do I communicate the story in my data?
Well, what is your purpose? Not all data visualizations are intended to produce the same type of “Aha!” moment in your audience.Understand that guidelines or rules are meant to help you communicate with your audience, but your intention or purpose of visualizing data will dictate whether or not those guidelines are worth following to communicate your story.
If your goal is to make beautiful art with data visualizations, then “aha” moment in your story is “look how beautiful this is”, not as much about the conclusions you can draw from the information itself. Other the other hand, you should approach your work differently if you are trying to nail your quarterly PowerPoint presentation to a board of directors.
If you’re trying to discover what’s important in your data, you will have a different purpose in storytelling than if you are trying to persuade others of what you already know to be important.
In “Storytelling with Data” Nussbaumer Knaflic delineates between these two purposes as “exploratory” vs “explanatory” analysis. She says “exploratory analysis is what you do to … figure out what … [is] noteworthy. [… Explanatory is when] you have a specific thing you want to explain”
We’ll start with Exploratory principles.
Lots of ways to do this, especially depending on your tool.
Generally though, I propose a more systematic approach, regardless of your data tool of choice. While its not a perfect approach, its a great way to familiarize your audience with your data – especially if you have the capability to expose data to them in a governed or data-discovery environment for exploration – (in other words, a dashboard). Dashboards can be good ways to put guardrails on user’s data exploration while still enabling them to find meaningful information.The Exploratory Design Principles I generally start with are:
Start with a summary
Allow for discovery with as much context and interactivity as possible
Verify with the details
“Summary -> Discovery -> Details”, or “Display -> Diagnose -> Decide”. Lots of catchy ways to say that, feel free to come up with your own snazzy alliterations.
Along with being a very logical approach to visualization, it also greatly reduces performance impact on analytic systems, since it should reduce the amount of data people return in their queries at any given time.
A great book I recommend for learning more how to do exploratory analysis is “Signal” by Stephen Few.
“Signal” talks a lot about how volumes and volumes of data require us to become adept at finding what is important. Lots of good stuff in that book, I encourage you to check that out if you’re looking for techniques and tips for understanding what matters in your data.
For explanatory analysis - To know how you should communicate, you start with what you need to communicate.
This is where people come in to the picture. Ultimately any data we have is lacking context to some degree, even in the age of information. Human intuition is invaluable to identify this gap of information – the trick is to allow data to challenge and confirm our intuitions. Another way I’ve heard this put is “Trust but verify”. We can use all sorts of analytics tools on the marketplace today to help our audience discovery what is important in their data. The secret sauce in your data is what you bring to the table. When you combine intuition with data, decision-making flourishes.
“Storytelling with Data” gives these helpful questions to ask before you attempt to persuade someone with data.
Who are you communicating with? Do they trust you?
What bias does my audience have? Will this make them supportive or resistant?What do you need your audience to know or do? Be willing to ask for action!
I love the NBA – ESPN does a great job of communicating what happened in a game using data visualizations that follow these guidelines to some degree.
Ultimately, the story conveyed might be different from game to game, but the visuals on their site when organized can give an audience a pretty good idea of the story arc of the game – and help people figure out what drove the outcome of the game.
Lets start with the way not to do it.
It is common to see tables in PowerPoint slides. Imagine you were trying to figure out what happened in the game last night. Looking at the box score can be a good start. You can tell who won, who contributed the most, you can see all the key metrics, and will get an idea for what was most important. But this forces people to read the data – to use their verbal system. What if we tried to communicate the game story using visuals or a dashboard instead?
What is most important is now front and center. Golden State won the game.
The most important visualization is in the top left, the score. The next tells the story arch of the game with the “Game Flow” chart. (side note - A great example of usage of this visualization translated in the business world could be actuals vs budget) – it shows when the game turned, it shows when teams went on a hot streak, and when they were unable to score.
The rest of the charts represent win probability, shot distribution, and key metrics comparisons. We’re focusing on what I think is most important, I’m not letting you pick and choose metrics, because my purpose is to EXPLAIN to you what happened, not leave much room for further exploration on info like usage rate or efficiency – even though those could be interesting Explorations.
In western language, we are trained to read things left to right, and top to bottom. Your data visualization should take advantage of this by ordering your visualizations by their importance in this way. I put the ESPN logo at the top right to convey to you that I did not make these visuals, and that information is important for me to convey. But not as important as who won the game.
Putting a logo in the top left might be a waste of space.
Lastly, Nussbaumer Knaflic in “Storytelling with Data” suggests you summarize when your purpose is explanatory - to visualize to persuade.
“If you had a single sentence to tell your audience what they need to know, what would you say?”Then, say it!
Then, include these tables at the back of your PowerPoint deck in an appendix for later reference.
Lets dive deeper into the data visualizations themselves now. Is there a way to determine effectiveness of data visualizations? Or is it merely subjective opinion for which charts are better than others?
We need to frame the debate here – effectiveness depends on the purpose of our visualizations, as we discussed. Since we marketed this webinar to business people and not art students though, I think it’s a fair assumption to make that we’re dealing with business people on this webinar.
My purpose is to help your organization translate your data into meaningful and actionable information. A great way to do this, and to measure effectiveness of visualizations, is to understand the cognitive load we place on our audience, and to know how to focus attention on what is important.
Let’s go rapid-fire through some examples
Which chart is easier to interpret quickly in exploratory analysis?
Simply put, our brains are better at seeing which values are greater than others in a simple bar chart vs a pie chart. If the goal is reducing cognitive lift, bar charts win the battle of effectiveness every time.
Which chart is clearer for explanatory analysis?
If you were to present this information to your audience, the chart on the left would leave no question which category you are planning to discuss. The chart on the right is noisy because of all the color.
Use color to draw attention to and emphasize what is most important, while still leaving context of the whole.
How about this one?
It depends! By connecting the dots, or not connecting the dots, you will take advantage of different pre-attentive attributes. The scatter plot in the top left would take advantage of proximity to define the relationship in the data by clustering, while the bottom right chart would use “connectedness” pre-attentive attribute to draw attention to the progression of information from left to right. Know your pre-attentive attributes, and choose the right one to convey relationships in your data. Don’t make your users have to think harder to undo a relationship you created by connecting the dots (or not connecting the dots). Either way, don’t choose scatter plots vs line charts flippantly.
Which value is most important in each of these visualizations?
It depends! Use sorting though to order visualizations by their values, and not their alphabetical name, to make it easier to read either way.
Better yet, combine principles to maximize the effect.
If my goal is to call attention to E being the lowest value, the bottom right chart is the best way to accomplish that in context of the other values –using color and sorting.
While we’re on this bar chart example, remember to include labels on your charts, especially in dashboard tools.It’s a pain to have to hover on each value to determine scale.
Or you can trade your axis lines for individual display values on the bars. Ultimately, determine your goal for this design decision. We’re in the subjective zone now, all of these are acceptable options.
Just be intentional with your choice
Bar charts are fine – but you didn’t think I’d only show bar charts in a data viz webinar, did you?Let’s talk about the concept of chart variety.Frankly, I see too much variety for the sake of variety in data visualizations today. We can get really exotic when it comes to presenting information, and provide a ton of value in doing so. Majority of the time though, most organizations can use these 12 visualizations to communicate your data - according to Knaflic in “Storytelling with Data”, these are her top 12.
Frankly, I’d eliminate square area from my “Top 12”.
The problem with the square area chart is, how do you know if the blue represents 25% of the data, or 20% of the data? In other words, does the gray extend behind the blue area, or are they separate. Our pre-attentive attribute of continuance is working against us here in my opinion.
One addition I would include in my top 12 are maps as well.
In terms of visualizations, I would recommend you keep a steady diet of these chart types when presenting information to people. Consider them your fruits and veggies, contrasted with “dessert” visualizations like pie charts, which hold some aesthetic value but ultimately fall short of solid analytic value, and end up making your audience work harder to understand the information presented.
Ok – Normally this is the point in the presentation where data viz experts climb onto their soap box and trash pie charts and tell you how any presentation with a pie chart loses their respect and attention.
This is not that presentation. My focus now is that “adoption is king”. “Adoption above all”
For those of you that don’t know – adoption means “people actually use this information – and buy into the way you’re communicating it”
Aesthetics matter. Visualizations referred to as “chart junk” matter. They just do. People get excited when they use a new visualization type to discover hidden information in their data, no matter how superficial it is.
Ultimately, the value of the information you provide is whether or not people consume it, and whether or not they do anything about it once they see it. Any way you can drive adoption, please do so with a clear conscious – even if that means throwing in a pie chart. Only, do so intentionally as a concession because your audience LOVES pie charts. Just remember to give them a steady diet of fruit and veggie visualizations (our top 12) vs just feeding them dessert.
So – in the name of adoption, lets get more exotic.
We’ll ease our way in here, less controversial - one important addition I would make to geo-mapping as a visualization type is mapping non-traditional areas, such as buildings, warehouses, and retail floors to unlock hidden relationships in your data. These are just sports examples from ESPN and StubHub, but I there are very cool examples from the business world as well.
Pareto charts (or cumulative percent of total charts) can help your audience focus on what is most important by showing percent of total. Use color to highlight the data above a certain % of total to focus attention on your categories that drive results. Careful not to include negative values here, it will lead to some odd behavior in your visualization axis.
I like this chart because it adds a light flare to the standard bar chart, and doesn’t exclude people from being able to derive information, even if they aren’t trained on how to read it.
Radar charts are an under-utilized way to compare multiple different categories against multiple different metrics.
One example would be measuring how well-rounded an employee is based on skill assessment data. With a radar chart, this representation of well-roundedness is quite literal, as the more shaded area an employee shows, the more well-rounded they are. This can also be used to identify skill gaps against multiple employees, as you can quickly see if one area of the web is lacking across multiple folks if they are sorted in the same order.
Box and whisker charts are a good way to identify the distribution of data, including outliers, mean, median, and quartiles. This is a tougher chart for some non-technical people to interpret, but it provides a wealth of information in a conscience, clear way. This one should be used only when training or explanations are provided to your audience.
A great place to go for inspiration on even more out-of-the-box visualizations is d3js.org, and their wiki gallery on GitHub. We’ll run through examples from those sites now. Most of them are very colorful and interactive as well – and can serve the purpose of art, exploration, or explanation depending on the chart.
Calendar visuals can be a great way to view seasonality in your data. This one by Mike Bostock and blocks.org helps us see week over week, month over month, and year over year changes in our data in an organized way.
Time-series visuals can be a great way to view similar information, but at an hourly level. This chart plots data at a minute level and includes one row per category or day. This could be used to analyze foot-traffic in a retail store location, or downtime in an industrial warehouse setting. You aren’t going to get very precise in your analysis here, but it gives you a high level “feel” for your data before diving into the details to understand specifics.
A simpler version of the same analysis could be to leverage a Gantt chart, like shown here, if a chart like the slide before is not available for you to use. It also takes out much of the noise to focus on what’s important – like “when were my machine downtimes yesterday”, vs showing all up-time and down-times.
The bullet chart can be thought of as a more successful, accomplished cousin to the gauge charts that are popular in analytics tools today. What I like about the bullet chart is that you can define a target, show the actual value, but then also provide supplementary information like quartiles, medians, or means to give context to the data as well.
Stephen Few loves these charts in “Information Dashboard Design”
Another one Mike Bostock put on d3js.org is the force-directed graph on the left. This highlights co-occurrence of characters in a novel, but you could imagine visualizing retail data for co-occurrence of products potentially as well here.
A similar approach would be to include a dependency wheel to show connectedness of your categories. The more connected they are, the wider portion of the circle a category is given here.
These aren’t very precise, and will likely require training again to avoid mis-communicating information. Again though, these are fun to look at, and if someone gets value out of this, they are going to feel like they found a needle in a haystack. Visualizing information in this way may lead to discoveries that would not have appeared as obviously in our top 12 charts, so be willing to think out of the box occasionally.
Notice how visually appealing these charts are too. Interactivity and color, while not particularly meaningful or strategic, will really stand out to your audience. If it is pleasant to look at, it is more likely to be paid attention to by your audience.
As a data visualizer and communicator, it is your job to manage the tension appropriately between aesthetics and efficiency and clarity. Once you know the rules, like pre-attentive attributes and the differences between your approach for exploratory and explanatory analysis, you can sleep well if you intentionally break them for the sake of adoption – as long as you do not misrepresent the information in your data. This balancing act is why data visualization will always be part-art, part-science, and part-politician.Manage the tension, and you will bridge the gap between your data and your audience.
In closing – remind yourself of these principles, and ask yourself these questions before you share data visualizations with others.
Thank you for joining our webinar today, I hope I’ve inspired you to go out and buy and read the data visualization books cited. There’s a wealth of information available today to help you, take advantage of them.
(We’ll be sending these out after the webinar as well)
In closing – remind yourself of these principles, and ask yourself these questions before you share data visualizations with others.