Infographics and Data Visualization

There are many ways to visually communicate variations in data sets. For example, weather maps use color to show differences in temperature, pressure, and precipitation across an area. Not every approach to data visualization is made equal. A landmark study by Cleveland and McGill (1980) found that viewers interpret some kinds of visualizations more accurately than others. They found that position (points on a graph or map) and length (such as in bar graphs) are the easiest to interpret accurately, whereas volume (increased size corresponding to increased number) and color (both hue and saturation) are the most difficult to interpret accurately. Here is a helpful chart we’ve created (based on a chart from an Intro Data Viz course) to demonstrate the ease of understanding:

Easiest to interpret

Hardest to interpret

This study and its evidence do not mean that you should use only bar charts, maps, or scatter plots to represent data. But if your data is complex and requires careful interpretation, consider using a visualization strategy that is easier to interpret, rather than relying on color or volume.

Pie and donut charts are everywhere! They visualize data by using angles (not area!) to represent parts of a whole. Generally, interpreting angles falls into the middle of the accuracy chart above. One reason is that there is no scale or axis to guide understanding.

Take a look at this pie chart to the right about our favorite pies.

At a glance, we can tell that the Chocolate pie is the smallest and the Apple pie is the largest. However, without additional information, we cannot accurately estimate the percentages for these sections. For example, the Apple pie appears to be about one-third (⅓) of the chart, and the Cherry pie is about one-quarter (¼). To accurately interpret the pie chart, we need percentage labels, which almost defeats the purpose of using a pie chart; there are better ways to represent this data visually.

In general, it is best to use a pie chart or donut chart only when representing two categories, one of which is larger than the other. That way, a viewer can quickly see which category makes up the majority, and the accuracy of interpreting the numbers is less important. If you have more than two categories, consider using another visualization approach, such as a horizontal stacked bar chart.

When representing our favorite pies in this chart, we don’t need numerical labels to interpret the data, as the axis labels (%) guide our understanding.

The stacked bar chart in the previous panel is easier to read accurately than a pie chart, but the visual noise does not help communicate the data. Let’s see how we might improve it further.

Even though it is not useful for accurately representing numerical differences, color is a really powerful tool for storytelling in data visualization. In the charts above, color is being used to distinguish categories. This is necessary because the category labels are in a key below the chart, rather than directly labeled in the charts. Although practical in that form, the colors don’t carry any real meaning in relation to the categories (in this case, types of pies) they represent and could potentially distract the audience from the actual information.

Let’s imagine that we want to tell a story about the most popular pie flavor. We can separate the different categories into individual bars and label the data directly. Then we can strategically use color to emphasize our argument that Apple pie is clearly the best pie. We can make a really strong visual argument by highlighting only Apple pie with color:

In the previous panel on the use of color, you’ll notice that the x-axis only goes up to 50% because none of our data points exceed 41%. As a result, the differences among our four pie categories are exaggerated. This can be a powerful tool for providing emphasis, but can also be considered misleading. If the absolute value of these favorite pies is important, we might want to set the x-axis to go from the minimum to the maximum possible value — in this case, 0% to 100%, as is shown below.

The downside to this type of representation is that there is now an excess of negative space. The question of scaling vs. absolute values often depends on the information presented and the norms of your field. From a design perspective, the scaled access is preferable because it fills the chart, but from an ethical data representation perspective, it may skew the understanding of the data.

A good data visualization uses ink or pixels sparingly. It is important to visually represent only what is necessary for the viewer to understand the information. Visual variation without meaning (eg., using different colors or line widths) can distract from the important information that you intentionally visualized. Or, even worse, it might lead viewers to draw the wrong conclusions. A busy visualization with too much variation muddies the message of the data.

Another thing to consider is what is necessary to include in a data visualization. Do you need axes, labels, etc. From a design perspective, reducing visual “noise” or things that can distract from the main information is key to quickly guiding the audience to understand your data visualization. For example, in our chart about favorite pies, do we even need an x-axis? With our data labels, it isn’t necessary, so we can remove it to reduce the visual noise.

The choice to remove noise or excess information will need to be guided by the medium in which the data is being visualized as well as the norms of your field. For example, if you were to put this chart on a PowerPoint slide, the stripped-down version above is better because you, as the presenter, can guide your audience to understanding without axis lines/scales. However, if this visualization were to be included in an academic paper, it might be better to include the axis lines since they will need to interpret the visual on their own.