How to design an effective data visualization in 5 steps?
In general, we tend to underestimate the value of proper data representation and often deliver quick, confusing and incorrect visualizations as a result. This note provides a high-level 5-step process that anyone can follow to create better visuals.
So why is data visualization important? Proper data visualization ensures that your message is not interpreted ambiguously, but effectively in a correct manner by your audience. When applied correctly, data visualization allows for your insights to be understood much quicker as humans are much better at interpreting visual cues in comparison to text based signals. While this note aims to be more practical and focuses on the process rather than the rationale, please check this article for more information on why data viz matters. In essence, every data visualization can be created effectively through 5 steps:
Start with your message! Before designing any visual, your message has to be absolutely clear. As a result, a user-centered approach is vital for any successful project. To do so, think about everything that the visual should convey and execute a requirements analysis with involvement of the assignment owner. Make sure to carefully document all of those requirements. Asking the following 5 questions often gets you a long way:
How will this visual be used?
Where would you access the visual?
Who will use the visual?
What is the key message of this visual?
What would you want users to take away from this visual?
After having clearly defined your message, you should involve your audience. Make sure you know who the people are that are going to use your visual and answer the following questions:
What is their proficiency with the subject matter that we’re talking about? (e.g. accounting metrics, operational quality numbers)
What is their data visualization proficiency?
Is this user group interested in detail and or a high level summary?
How frequently will my audience use this visual? (e.g. is this a one-time presentation or will this be part of a standard excel file, website or dashboard)
Once you have a better overview of your audience, start designing with their needs as well as the needs of the assignment owner in mind.
When your assignment owner, your audience and yourself are aligned, you can start thinking about the best possible medium for your message. There is no one-size-fits-all solution for your data visualization. Consequently, it’s vital that you choose one that works for your situation. You can do so by employing a chart guide like the one listed here. A chart guide is an easy-to-use tool for getting a sense of which visualizations work best for certain use cases. Let’s use the following two scenario’s to illustrate:
Scenario 1: Your visual will present data on daily volumes per abuse type. Your audience consists of a team of technical specialists who need to see the data dissected by many different categories. In this instance, your visual is supposed to contain a lot of information and your audience is a small set of technical people. Here a nicely formatted table can be more than sufficient to present the data set for further processing:
Scenario 2: Your visual will present data on monthly volumes for one abuse type. Your audience consists of a team of managers who need a high level overview to pinpoint problem areas. In this scenario, your audience is a group of people wanting to have high-level insights on abuse volumes. Since the data set is limited, maybe a simple trend graph is a better option to quickly highlight outliers.
Usually, a column chart, bar chart, line chart and table will be sufficient for most of your visualization scenario’s. While visually appealing, a pie chart is one of the worst visuals to use. This is due to the fact that the human brain can fail to correctly estimate quantities and rankings due to perception of the size of the slices, as mentioned in this article. As a result, try to steer yourself to proven best-practices.
4. Visual Elements
After selecting your format (e.g. I’m going to use a scatter-plot to visualize my survey response scores, because I think the different patterns are most suitable in that way) you can think of which elements you want to use to highlight and set aspects of your data apart. For example, I’m working on sales data and I have chosen a bar chart for my data. Which elements do I want to use to showcase that category A has made less revenue than category B? Color could be used to so, but so could the length of the bar or the thickness of the bar. The following two examples showcase how the same data-set can be presented in the same chart, but employing different visual elements.
The example at the top uses the length of the bars to convey the amount of total and active users. Both metrics are grouped per province. The example at the bottom takes a different approach and uses length only for the total amount of users. Here, the active users are indicated through a color scale.
While both in essence present the same data. The usability of the two graphs differs severely. The example at the top is in most scenario’s easier to digest as the reader doesn’t have to connect the different color scales to the bars and can read of the exact number instead of a range as defined through 4 different categories. Consequently, knowing which elements work is vital. Some elements work best to derive precise insights like length and thickness, but some elements merely catch attention, like color. In addition, many people suffer from some sort of color-blindness, thus preventing the usefulness of colors to make precise distinctions. For an overview of visual elements and their effectiveness, please see this image on chart choice in this article. As mentioned in the article, there are a range of elements to choose from:
Grouping: Grouping items together to show patterns and trends
Length: Elongating parts to emphasize a difference
Thickness: Using the thickness of an element like a bar to visualize a difference
Light: Using the way an element is illuminated / shaded to highlight a difference
Color: Employing a color to visualize differences
Shape: Using a shape to differentiate
Size: Using the size of an element to showcase a distinction
Alignment: Aligning objects to highlight a difference
The last, but maybe the most vital, point is around the lay-out of your visual. In most cases, a visual that looks bad will not be used effectively or will even not be used at all. For that reason, the choices made here are very important if you want to emphasize the correct insights from your data. So what do we mean with lay-out? Lay-out ties in closely with your visual elements. However while the visual elements piece tries to explain what elements you should use, the lay-out explains how these elements should be formatted. This last step is one that’s often overlooked, but can really make or break your visual. Ask yourself which fonts, colors, titles, axes and all other texts should be optimized to make sure that your insights standout. Please see the following golden rules for your lay-out:
Stick to one light font: Multiple and or heavy distracting fonts can take away from your visual. Use only one simple font that’s light on the eye like ‘Arial’, ‘Segoe UI Light’ or ‘Verdana’.
Use pastels instead of saturated colors: Saturated colors look very harsh on the eyes and often draw unwanted attention. Often you don’t need to use the brightest red to convey that something needs attention.
Use clear titles: A title should be descriptive and aid the user in understanding your insights. ‘Metric 1’ is much less descriptive than ‘False positive rates for Graphic Violence in Q3 2020’.
Limit your axes: Use axes only when you’re not employing data labels and vice versa. Redundancy can clutter up your visual. Also make sure to always start your axis at 0 to avoid portraying a wrong message.
Eliminate non visual text: Text that is not essential to portray your data only distracts from your message. So make sure to eliminate or minimize the use of logos, background images and all other forms of clutter if they’re not needed.
Please see the images above as an example. While the top image incorporates the right format and visual elements, the lay-out of the visual makes it hard to interpret. A cleaner look was achieved by changing the following:
Sorting the bar chart from largest to smallest so it’s easier to spot outliers.
Removing the different colors as they don’t represent any data point.
Removing the drop shadow from the bars for decoration.
Removing the grid-lines and x-axis as the bars already have data labels, making them redundant.
Truncating the data labels to 1 decimal.
Using a lighter more neutral font for the axis, data labels and title.
Removing the decorative background image and logo.
Adding a more descriptive title.
In short, the importance of data visualization should not be underestimated. For a maximum impact on your users, make sure that you put meaning into your visuals while presenting them in a bite-size manner. To do so, use methods like this very simple 5-step approach to guide you through the process.
Learn more on applying much of the same principles to designing dashboards through this great note by Sumit: https://fb.workplace.com/notes/sumit-sharma/the-5-key-dashboard-design-steps-with-data-visualization-best-practices/249047989767148/ Article on the importance of data visualization and the use of visual elements: https://www.knowablemagazine.org/article/mind/2019/science-data-visualization Chart choice graph: https://github.com/ft-interactive/chart-doctor/tree/master/visual-vocabulary Article covering importance of data visualization: https://hdsr.mitpress.mit.edu/pub/zok97i7p/release/2 Data visualization e-learning (Adobe Illustrator): https://facebook.csod.com/ui/lms-learning-details/app/course/b9f5d524-22c5-5e22-b273-a0d93b05f18d Storytelling with data e-learning: https://facebook.csod.com/ui/lms-learning-details/app/course/015a0b4a-8edb-5dfc-8553-bc09ddddd095?isCompletionRedirect=true&loStatus=16®num=1&loId=015a0b4a-8edb-5dfc-8553-bc09ddddd095 Why pie-charts are the worst: https://www.businessinsider.com/pie-charts-are-the-worst-2013-6?r=US&IR=T