“Unless we know what we’re doing, we’ll end up with a cluttered mess.” These are wise words from one of the most widely known data visualization experts, Stephen Few, in his book, Information Dashboard Design: Displaying Data for At-A-Glance Monitoring (2nd edition, p. 93). This gets to the heart of why it’s important to spend the time understanding and implementing data visualization best practices.
Entire books have been written on this subject. What I seek to produce here is a condensed version of what I believe to be the essential points for ensuring your visuals, reports, and dashboards are good, clean, and informative, rather than bad, cluttered, and nonsensical. Data visualization is both an art and a science. The principles of good data visualization design make up the science, whereas the application of the principles in various contexts and with various styles makes up the art. Like any art form, it takes practice to get really good. Learning is an iterative process, so iterate away!
Tip #1: Choose the right chart/graph. Choosing the right visual for the job (the type of data you’re analyzing and what you want to communicate about it) is a critical first step. Ask yourself the question, “What do I want to show?” There are four main answers to this question: comparison, relationship, composition, and distribution. You can use the chart below, by Dr. Andrew Abela, as a guide.
Keep in mind that not all data visualization types are created equal and the one you choose should be influenced by the context of your analysis and audience.
To pie or not to pie? The use of pie charts and donut charts is somewhat of a debate in the data visualization community. They are very common in business contexts; however, most data visualization professionals will quickly tell you to avoid them as they’re difficult to interpret. Take the below image as an example (from Wikipedia). First, try to compare size of pie slices within each pie. Then try to compare the same colored pie slices between A, B, and C. It’s challenging, isn’t it? Now, do the same thing with the column charts. Much easier, right?
As you can see, it’s easier to interpret length than it is to interpret curviture and area. Now, I’m not going to tell you that you should never use pie and donut charts. That would be hypocritical of me. There are rare cases when I use them, but they must be used properly. If you do use a pie or donut chart, you should limit the number of categories they display to about 2 or 3. Below is an example of a big no, no.
I’m ashamed to say it, but I used to do this, thinking it looked so cool with all the colors. If only my future self could go back in time to slap my past self upside the head. Now, here are two examples of properly utilizing these dessert charts.
Note how clean and simple these charts look. You can actually gain insight from them with little to no difficulty. But what if you need to visualize a lot of categories, like in the pie and donut charts above this one? A good alternative to pie/donut charts is a bar chart, sorted in descending order.
Tip #2: Choose colors wisely and sparingly. The less the better. I’ve seen visuals that have almost a rainbow color palette, like the pie and column charts from Wikipedia above (yuk!), which is just distracting and gaudy. Don’t be gaudy. Be clean and minimalist. In her book, Storytelling with Data: A Data Visualization Guide for Business Professionals, Cole Nussbaumer Knaflic provides valuable insight into color utilization:
For color to be effective, it must be used sparingly. Too much variety prevents anything from standing out. There needs to be sufficient contrast to make something draw your audience’s attention.Ch. 4 Focus Your Audience’s Attention; Color; para. 4
In other words, don’t go wild with colors. I’ve noticed that new-comers to data visualization tend to overuse color in their visuals and reports, assuming that more color equals “looks more interesting”. This, however, is typically not the case. And even if a data viz did look more interesting as a result of more colors, that doesn’t equate to more informative and understandable, which is ultimately what you should be aiming for. I’ve seen really interesting, informative, and understandable data visualizations in grayscale. Yes, grayscale. So think twice before splashing more color onto your data viz. Ask yourself, “Is it necessary?”
Lastly, you usually want to use muted/soft colors, as opposed to bright or bold colors which tend be harsh on the eyes.
Tip # 3: Label, Label, Label. What does the data viz show? Put it in context. People won’t know unless you have a title, possibly a subtitle, and your axes are properly labeled. Additional labeling of key data points can also be utilized to emphasize a point and tell a story. It may also be a good idea to provide a summary of findings from the data viz, making it easier for people to understand. This essentially makes your data viz serve as evidence of what you’re communicating. Here is a data viz I put together for #MakeoverMonday for the 52nd week in 2018.
Note how I use a question as my title to help the audience think through the subject at hand. I provide a summary of the data and my findings just below the title. Additionally, I provide supporting text in the body of the data viz to emphasize key points and tell a story. Lastly, the axes are properly labeled. You’ll also notice that I utilized the tip on color choice: choose wisely and sparingly.
Tip #4: Choose a proper layout. This tip kind of has more application for reports and dashboards than individual visuals, but it still applies. For instance, alignment falls under layout and you want to make sure your title is left-aligned and any labels you have in the visual are aligned in such a way that they don’t clutter the view.
When it comes to reports/dashboards, you want to have higher level visuals (low granularity) at the top and lower level visuals (high granularity) at the bottom. In other words, visuals that contain highly aggregated views of the data should be at the top, whereas other visuals should be at the bottom. You also want the more critical data points or metrics in the upper-left corner, as this is usually where people look first (just like reading a book, left-to-right). This aids in the interpretive process. Following are some examples of report layout best practices.
Tip #5: Think outside the box. Be an out-of-the-box kind of thinker. Sometimes the typical chart and color scheme approach is OK. Other times you may need to go the extra mile to have that powerful impact and communicate a story. Below is a data viz I developed in Power BI that falls under the category of thinking outside the box. This viz shows NBA fan cost (cost of tickets, food, apparel) for different NBA teams. I developed the visual in such a way that it actually looks somewhat like two basketball teams getting ready to face off. The team slicers (top and bottom) look like the team benches. This was developed using scatterplots, data cards, shapes (the lines), text, and slicers.
Tip #6: Think like a Data Analyst. Visualizing data necessarily involves analyzing data. This involves understanding different data types, statistical concepts, how to transform data, and perform calculations. The more proficient you are at these things, the easier data visualization will be. Additionally, you’ll be able to think through more advanced ways to visualize the data and be able to communicate more advanced statistical findings from the data.
Please understand, you don’t have to be statistician or have an advanced degree in mathematics. Even grasping fundamental statistical concepts such as the central limit theorem and the basics of inferential statistics can go a long ways. If you lack experience in this area, it’s a good idea to take some time to get familiar with it. It will only benefit you in the long run. I highly recommend checking out available courses on Udemy in probability and statistics.
Summary. I hope you found these quick tips helpful. If you start implementing these principles of data visualization best practices then your visuals will be cleaner, more appealing, easier to understand, and more impactful. That being said, these tips just scratch the surface. If you’re looking for a regular challenge in this area then I encourage you to join the #MakeoverMonday community project. Not only will you be challenged to work on unfamiliar data sets and put these principles to good use, but you’ll have lots of opportunities to learn from others in the community.
Additionally, there are lots of good books available on this subject. Following are just a few that I recommend.