Choosing Chart Choosers (with Nick Desbarats)

Talking about Chart Choosers with Nick Desbarats

A couple of weeks ago, I posted an illustration on LinkedIn that was almost a provocation. It was titled “Same data, different charts,” and I added the image below:

This is what I wrote:

One of the biggest unsolved challenges I see in visualization is having better models to think about what makes one representation *of the same data* preferable to another. The sketch below gives a tiny glimpse of what I am talking about. This is the same data represented with a (very small) set of alternatives. There are way more! How is one supposed to choose between so many? Well ... first, one needs to be trained in producing so many options, but then one needs some criteria to help choose solutions. The best we have is this vague idea that some "visual channels" are better than others, but it's very limited. What else is there to help with this choice? I don't have good answers yet. Any thoughts???

The post sparked many reactions, and among the many conversations, I was intrigued by the replies from Nick Desbarats. Nick is an independent educator and book author who, among many other things, developed chart choosers that look like no other. If you are not familiar with them, chart choosers are tools that help people choose the “right” chart according to some criteria. Most chart choosers suggest appropriate charts according to what kind of data one wants to visualize. Still, some others use different criteria, such as the communicative intent of the visualization. My favorite one of these second types of choosers is the Financial Times’s “Visual Vocabulary,” which suggests appropriate charts for whether you want to communicate a deviation, correlation, ranking, etc.

In any case, after a few back-and-forths, Nick and I decided to jump on a call to continue our conversation and record it to make it public for our respective audiences. You can watch our 20-minute chat in the video below. In the following, I’ll summarize a few of the main points I found particularly interesting.

(Side note 1: My recording is, unfortunately, a bit out of sync. Our recording platform had a glitch, and we could not fix it. I hope you’ll enjoy it anyway. I am still learning how to do video recording with guests properly!)

(Side note 2: I now have a YouTube channel! Yes … with one, just one video for now. 🤷‍♂️ But I will certainly keep adding new videos there as I ideate more video content. Subscribe to the channel if you are interested!)

Problems with existing choosers

Both Nick and I agree that the problem with existing chart choosers is that they are either based on data or purpose, and both approaches have limitations. Those based on data are fundamentally underspecified or may even lead to the wrong decisions. For example, Nick said:

“Oh, you know, you're showing the breakdown of a total? Well, you can use a pie chart or a stacked bar chart, or regular bar chart or waterfall chart, you figure it out.”

This is clearly a problem if you do not have other criteria to choose among the available choices.

I like this specific quote from Nick, which should probably be printed and hung on the wall:

What is the best way to show this data? Is the wrong question.

This is something I keep telling my students and an aspect of data visualization I have found frustrating for a long time.

Again, Nick elaborates on that later on:

When you create a chart […] Do I know why I'm creating this chart? Yeah, what question am I trying to answer? What problem are they trying to illustrate? Because if you don't know that, if all you have is data, but you don't know what you're trying to say about the data, well, then you won't be able to answer any of these questions.

However, limitations also exist for chart choosers based exclusively on purpose. Later on, Nick states:

But then there are others, which depend entirely on the purpose of the chart. And so you have to basically figure out, what am I trying to communicate here? What is the question, the specific question I'm trying to answer? What is the kind of comparison that I'm trying to make? What is the particular problem that I'm trying to illustrate?

And, of course, this means that if you do not have a way to choose according to purpose, you won’t be able to make good decisions.

Who needs chart choosers?

I was struck by Nick's brief comments regarding who needs chart choosers. It’s quite obvious that a super-experienced visualization designer is not the right target for chart choosers. In fact, these people may very well end up even breaking the “rules” of a chart chooser because they know better than any algorithmic approach to the problem of choosing the right visual representation.

So, chart choosers are for inexperienced people, and many are not particularly interested in visualization; they just want to get the job done! As someone so immersed in studying this idea, I tend to forget that many people out there need our help but do not care about the intricacies of our field; they just want to solve a specific problem they have. I think this is an important lesson to always keep in mind.

I want to add, even if I did not mention this in the video, that chart choosers also have a pedagogical value. I started using them a bit in my classes. I think they help expose students to problems and inconsistencies, so I believe they are also useful for this specific purpose, not only to make the choices they are designed for.

Scalability matters

When Nick started showing his chart choosers, I noticed that some choices he includes pertain more to scalability aspects of the data than what type of data one has, and I commented on it by saying:

What I noticed in my own practice is that sometimes I have to switch from one plot to another just because one is not scalable enough. I think a classic example would be […] switching from a bar chart to a tree map.

This is an incredibly overlooked aspect that I have seen captured only in Nick’s chart choosers. Often, when we are confronted with the problem of visualizing a given data set, choosing between one representation or another is a matter of scalability. Nick explains it very well here by giving one example:

You know, if you have cyclical data, you know, seasonal data, for example, or weekly cycles, one of the questions is, how many cycles are there? Because if there are fewer than six or seven, you can actually show those as an overlapping line chart. […] But if you have more than six or seven, this is going to be a spaghetti mess and so then you have to switch to […] a heat map.

Scalability is a huge topic that is often neglected and has implications regarding the idea of visualizing data using the most effective visual channels. There are situations like the one illustrated by Nick above where visualizing quantitative values with color intensity, for example, makes total sense even if visualizing quantities with color is, in general, a poorer choice than, say, using size.

Being specific with purpose

Initially, I wrote that some chart choosers are organized around purpose. I like this type of chart choosers, but the effectiveness of decisions based on purpose also depends on how we define purpose and how granular these decisions are. In Nick’s charts choosers, I also like how specific the decisions based on purpose are. Here is one of my comments I had while exploring one of his chart choosers:

So what I what I see here, for example, you have: is it more important to show subtotals or fraction of total? […] So you can answer these only if you know […] what's the purpose, but this question is very specific. So it's walking you through, making purpose way more concrete.

I think there is something to this idea of codifying purpose better and seeing how a more granular characterization could help us make more mindful choices when we design a data visualization.

Pie charts are fine!

I can’t recall if I explicitly stated this in the newsletter before, but I am not against pie charts. I think they have their place and work well when used appropriately. The problem is knowing when and how to use them appropriately! This is a clear point Nick makes. Pie charts are not universally bad or good; they are good in specific cases, and if you don’t know what these specific cases you may end up with suboptimal results. I like what Nick said regarding the overgeneralization of the problem with pie charts:

Then people kind of overgeneralize and they come to the wrong conclusion. Like they were like, oh, you know, this pie chart is is terrible. It's useless. And so we shouldn't use pie charts.

This is an excellent point that could be extended to many other cases in visualization. For example, I find a few charts quite unnecessarily complex and confusing (like connected scatter plots) that I am sure have some good use cases. The point is to learn what these specific use cases are.

Conclusions

While there are many chart choosers, there is surprisingly little research on their differences and commonalities, as well as critical reflections on their premises, design, and user. As I wrote this, I realized that chart choosers could be an interesting area of research for data visualization.

Chatting with Nick has been great, even if our chat lasted only about 20 minutes. His work made me reflect more deeply on how we make choices when we design data visualizations.

I’ve also enjoyed working on this specific format. It’s interesting how this newsletter, social media sites like LinkedIn, and video or audio recording can synergistically play a role. I think I’ll try to experiment more in the future. Being able to jump on a quick chat with someone, record it, and publish it seems to be a really good model for this newsletter.

—

What do you think?

Do you like this format? I know it can be improved, but I would love to get a general sense from my readers if they like the format now or not. Can you please let me know by writing a comment below? Thanks!