Data Ab Initio

I’m starting to do more work in the area of data visualization, as evidenced by my most recent publication and my somewhat periodic Twitter rants against pie charts. While I’ve already discussed one of my favorite visualization resources on the blog, I have not discussed how I approach data visualization. Particularly, I want to share the framework I teach in my visualization workshops.

The framework involves three steps:

1. Determine your message
2. Determine what type of data this corresponds to
3. Choose the best visualization format to match the message and data type

Let’s break each step down in more detail.

The first step in any visualization is determining what you want say. While you can just visualize all of your data and let the reader decide, this makes more a lot work for the reader; they’ll either give up or take time away from understanding some other key point in your work to understand the data in this visualization. To illustrate why message is important, consider Nathan Yau’s exercise in visualizing the same data 25 different ways. Each visualization highlights a different nuance of the data and causes the viewer to interact with the data differently. So the message is key to determining which is the “best” in the set of 25. In your figures, you can both be transparent by visualization everything and frame things in a way to make a point. Basically, having a message can really help you add clarity to a figure.

Once you determine your message, you’ll next want to consider what type of data you’re working with. For example, while you might have survey results, depending on your message you could really be trying to visualize a single number, a comparison, a change over time, etc. Each of these data types merits a slightly different visualization strategy. So spend some time thinking about your message and how it directs the specific data features you want to highlight.

The final step in this framework is to actually chose the type of visualization or chart to use. By making this decision last, you can reflect on how the chart type best lines up with what you’re trying to say. As you pick your chart type, continually ask if this visual fits your message. If it doesn’t, pick a different chart type.

I fully admit to building this framework off the work of Stephanie Evergreen. If you want some structure for determining the best chart types for different data types, I recommend checking out her book “Effective Data Visualization”. For qualitative data, she also has a qualitative chart chooser on her website. These resources will help you make the jump from data types to chart types.

I should also note that this framework is for visualizing final results. If you’re looking to do visualization for exploratory analysis, that’s a totally different topic and one with other resources (e.g. Tukey’s “Exploratory Data Analysis” or R’s summarytools package).

So that’s one way to start thinking about making good visualizations. If you’re going to go to the effort to visualize your data, I challenge you to make your visuals their most effective by determining your message and the key data features before you decide on the chart type.

Today on the blog, I’m going to talk about the difference between a bar chart (right-left bars) and a column chart (up-down bars). For many, the difference seems negligible — simply the direction of the bars — but the choice between the two can often make a difference in terms of readability of your chart.

I’ll demonstrate my point using a recent published example of a column chart that might be better as a bar chart. The paper the chart comes from is really great and my critique of its Figure 6 in no way means that I don’t find the rest of the article to be of excellent quality. The chart is one of several examples I’ve seen but I’m allowed to reproduce it here under a CC BY license.

Here is the column chart in question. It displays where data curation actions are happening and overlays people’s satisfaction with that activity. Can you spot the major issue?

I’m really interested in the findings in the chart but I can’t get much actionable information out of it because so few of the columns are labelled (or even fully labelled). Additionally, it’s difficult to match the existing labels to columns due to the diagonal orientation of the text. All this is an artifact of the charting software compressing labels to fit in the allotted space.

The key point I want to make is that label dropping is less likely to happen in a bar chart (right-left bars). It’s usually easier to make a chart taller than to make it wider in order to provide the necessary label space. Plus, the text automatically orients itself in the proper direction so you don’t need to turn your head to read the chart. Overall, this makes for a more readable, more usable chart.

To show the difference, I reproduced a subset of the column chart as a bar chart. I think that it’s much easier to read and take away concrete findings because you can interact with every category in the chart.[1]

So next time you’re making a column chart that needs a lot of labels, I hope you consider using a bar chart instead. It really does make a difference and your readers will appreciate a more readable chart.

 

 

[1] There’s an artifact in Excel not letting me use an overlay line as in the original chart but I think the overlapped bars sufficiently demonstrate the original point.