Storytelling with Data: A Data Visualization Guide for Business Professionals

by Cole Nussbaumer Knaflic

Use cases for stacked vertical bar charts are more limited. They are meant to allow you to compare totals across categories and also see the subcomponent pieces within a given category.

Figure 2.14 Bar width

Figure 2.16 Comparing series with stacked bar charts

When you use the 100% stacked bar, think about whether it makes sense to also include the absolute numbers for each category total (either in an unobtrusive way in the graph directly, or possibly in a footnote), which may aid in the interpretation of the data.

The best way to illustrate the use case for a waterfall chart is through a specific example. Imagine that you are an HR business partner and want to understand and communicate how employee headcount has changed over the past year for the client group you support. A waterfall chart showing this breakdown might look something like Figure 2.17. Figure 2.17 Waterfall chart On the left-hand side, we see what the employee headcount for the given team was at the beginning of the year. As we move to the right, first we encounter the incremental additions: new hires and employees transferring into the team from other parts of the organization. This is followed by the deductions: transfers out of the team to other parts of the organization and attrition. The final column represents employee headcount at the end of the year, after the additions and deductions have been applied to the beginning of year headcount.

Brute-force waterfall charts If your graphing application doesn’t have waterfall chart functionality built in, fret not. The secret is to leverage the stacked bar chart and make the first series (the one that appears closest to the x-axis) invisible. It takes a bit of math to set up correctly, but it works great. A blog post on this topic, along with an example Excel version of the above chart and instructions on how to set one up for your own purposes can be downloaded at storytellingwithdata.com/waterfall-chart.

Horizontal bar chart If I had to pick a single go-to graph for categorical data, it would be the horizontal bar chart, which flips the vertical version on its side. Why? Because it is extremely easy to read.

The horizontal bar chart is especially useful if your category names are long, as the text is written from left to right, as most audiences read, making your graph legible for your audience. Also, because of the way we typically process information—starting at top left and making z’s with our eyes across the screen or page—the structure of the horizontal bar chart is such that our eyes hit the category names before the actual data. This means by the time we get to the data, we already know what it represents (instead of the darting back and forth our eyes do between the data and category names with vertical bar charts).

Like the vertical bar chart, the horizontal bar chart can be single series, two s...

The logical ordering of categories When designing any graph showing categorical data, be thoughtful about how your categories are ordered. If there is a natural ordering to your categories, it may make sense to leverage that. For example, if your categories are age groups—0–10 years old, 11–20 years old, and so on—keep the categories in numerical order. If, however, there isn’t a natural ordering in your categories that makes sense to leverage, think about what ordering of your data will make the most sense. Being thoughtful here can mean providing a construct for your audience, easing the interpretation process. Your audience (without other visual cues) will typically look at your visual starting at the top left and zigzagging in “z” shapes. This means they will encounter the top of your graph first. If the biggest category is the most important, think about putting that first and ordering the rest of the categories in decreasing numerical order. Or if the smallest is most important, put that at the top and order by ascending data values.

Stacked horizontal bar chart Similar to the stacked vertical bar chart, stacked horizontal bar charts can be used to show the totals across different categories but also give a sense of the subcomponent pieces. They can be structured to show either absolute values or sum to 100%. I find this latter approach can work well for visualizing portions of a whole on a scale from negative to positive, because you get a consistent baseline on both the far left and the far right, allowing for easy comparison of the left-most pieces as well as the right-most pieces. For example, this approach can work well for visualizing survey data collected along a Likert scale (a scale commonly used in surveys that typically ranges from Strongly Disagree to Strongly Agree),

Figure 2.19 100% stacked horizontal bar chart

I avoid most area graphs. Humans’ eyes don’t do a great job of attributing quantitative value to two-dimensional space, which can render area graphs harder to read than some of the other types of visual displays we’ve discussed. For this reason, I typically avoid them, with one exception—when I need to visualize numbers of vastly different magnitudes. The second dimension you get using a square for this (which has both height and width, compared to a bar that has only height or width) allows this to be done in a more compact way than possible with a single dimension,

Figure 2.20 Square area graph

There are also some specific graph types and elements that you should avoid: pie charts, donut charts, 3D, and secondary y-axes.

now I’ll articulate a relevant data visualization rule: don’t use 3D! It does nothing good, and can actually do a whole lot of harm, as we see here with the way it skews the visual perception of the numbers.

Even when we strip away the 3D and flatten the pie, interpretation challenges remain. The human eye isn’t good at ascribing quantitative value to two-dimensional space. Said more simply: pie charts are hard for people to read. When segments are close in size, it’s difficult (if not impossible) to tell which is bigger. When they aren’t close in size, the best you can do is determine that one is bigger than the other, but you can’t judge by how much. To get over this, you can add data labels as has been done here. But I’d still argue the visual isn’t worth the space it takes up.

One approach is to replace the pie chart with a horizontal bar chart, as illustrated in Figure 2.23, organized from greatest to least or vice versa (unless there is some natural ordering to the categories that makes sense to leverage, as mentioned earlier). Remember, with bar charts, our eyes compare the end points. Because they are aligned at a common baseline, it is easy to assess relative size. This makes it straightforward to see not only which segment is the largest, for example, but also how incrementally larger it is than the other segments.

bar. The unique thing you get with a pie chart is the concept of there being a whole and, thus, parts of a whole. But if the visual is difficult to read, is it worth it? In Figure 2.23, I’ve tried to address this by showing that the pieces sum to 100%. It isn’t a perfect solution, but something to consider.

If you find yourself using a pie chart, pause and ask yourself: why? If you’re able to answer this question, you’ve probably put enough thought into it to use the pie chart, but it certainly shouldn’t be the first type of graph that you reach for, given some of the difficulties in visual interpretation we’ve discussed here.

Don’t use donut charts.

One of the golden rules of data visualization goes like this: never use 3D. Repeat after me: never use 3D. The only exception is if you are actually plotting a third dimension (and even then, things get really tricky really quickly, so take care when doing this)—and you should never use 3D to plot a single dimension.

Secondary y-axis: generally not a good idea

you should avoid the use of a secondary or right-hand y-axis. Instead, think about whether one of the following approaches will meet your needs: Don’t show the second y-axis. Instead, label the data points that belong on this axis directly. Pull the graphs apart vertically and have a separate y-axis for each (both along the left) but leverage the same x-axis across both.

Figure 2.27 Strategies for avoiding a secondary y-axis

third potential option not shown here is to link the axis to the data to be read against it through the use of color. For example, in the original graph depicted in Figure 2.26, I could write the left y-axis title “Revenue” in blue and keep the revenue bars blue while at the same time writing the right y-axis title “# of Sales Employees” in orange and making the line graph orange to tie these together visually. I don’t recommend this approach because color can typically be used more strategically.

It is also worth noting that when you display two datasets against the same axis, it can imply a relationship that may or may not exist. This is something to be aware of when determining whether this is an appropriate approach in the first place.

When you’re facing a secondary y-axis challenge and considering which alternative shown in Figure 2.27 will better meet your needs, think about the level of specificity you need. Alternative 1, where each data point is labeled explicitly, puts more attention on the specific numbers. Alternative 2, where the axes are shown at the left, puts more focus on the overarching trends. In general, avoid a secondary y-axis and instead employ one of these alternate approaches.

If you’re wondering What is the right graph for my situation?, the answer is always the same: whatever will be easiest for your audience to read.

There is an easy way to test this, which is to create your visual and show it to a friend or colleague. Have them articulate the following as they process the information: where they focus, what they see, what observations they make, what questions they have. This will help you assess whether your visual is hitting the mark, or in the case where it isn’t, help you know where to concentrate your changes.

every single element you add to that page or screen takes up cognitive load on the part of your audience—in other words, takes them brain power to process. Therefore, we want to take a discerning look at the visual elements that we allow into our communications. In general, identify anything that isn’t adding informative value—or isn’t adding enough informative value to make up for its presence—and remove those things.

designers of information, we want to be smart about how we use our audience’s brain power. The preceding examples point to extraneous cognitive load: processing that takes up mental resources but doesn’t help the audience understand the information. This is something we want to avoid.

This can also be referred to as maximizing the signal-to-noise ratio (see Nancy Duarte’s book Resonate), where the signal is the information we want to communicate, and the noise are those elements that either don’t add to, or in some cases detract from, the message we are trying to impart to our audience.

What matters most when it comes to our visual communications is the perceived cognitive load on the part of our audience: how hard they believe they are going to have to work to get the information out of your communication. This is a decision they likely reach without giving it much (if any) conscious thought, and yet it can make the difference between getting your message across or not.

In general, think about minimizing the perceived cognitive load (to the extent that is reasonable and still allows you to get the information across) for your audience.

clutter. These are visual elements that take up space but don’t increase understanding.

There is a simple reason we should aim to reduce clutter: because it makes our visuals appear more complicated than necessary.

When our visuals feel complicated, we run the risk of our audience deciding they don’t want to take the time to understand what we’re showing, at which point we’ve lost our ability to communicate with them. This is not a good thing.

Proximity We tend to think of objects that are physically close together as belonging to part of a group. The proximity principle is demonstrated in Figure 3.1: you naturally see the dots as three distinct groups because of their relative proximity to each other.

Figure 3.1 Gestalt principle of proximity We can leverage this way that people see in table design. In Figure 3.2, simply by virtue of differentiating the spacing between the dots, your eyes are drawn either down the columns in the first case or across the rows in the second case.

Similarity Objects that are of similar color, shape, size, or orientation are perceived as related or belonging to part of a group. In Figure 3.3, you naturally associate the blue circles together on the left or the grey squares together on the right.

Figure 3.3 Gestalt principle of similarity This can be leveraged in tables to help draw our audience’s eyes in the direction we want them to focus.

In Figure 3.4, the similarity of color is a cue for our eyes to read across the rows (rather than down the columns). This eliminates the need for additional elements such as borders to help direct our attention.

Enclosure We think of objects that are physically enclosed together as belonging to part of a group. It doesn’t take a very strong enclosure to do this: light background shading is often enough,

Figure 3.5 Gestalt principle of enclosure One way we can leverage the enclosure principle is to draw a visual distinction within our data,

Closure The closure concept says that people like things to be simple and to fit in the constructs that are already in our heads. Because of this, people tend to perceive a set of individual elements as a single, recognizable shape when they can—when parts of a whole are missing, our eyes fill in the gap.

Figure 3.7 Gestalt principle of closure It is common for graphing applications (for example, Excel) to have default settings that include elements like chart borders and background shading. The closure principle tells us that these are unnecessary—we can remove them and our graph still appears as a cohesive entity. Bonus: when we take away those unnecessary elements, our data stands out more,

Figure 3.8 The graph still appears complete without the border and background shading

Continuity The principle of continuity is similar to closure: when looking at objects, our eyes seek the smoothest path and naturally create continuity in what we see even where it may not explicitly exist.

stripping away unnecessary elements allows our data to stand out more.