16  Faceting aka Layout

The “faceting” feature of CODAP is worth a chapter of its own. I’m using this f-word in this book partly as a sop to users of R and ggplot, because that’s what it’s called there. But in the CODAP interface, you’ll see the word “Layout” in tool tips.

16.1 Introduction

But what is it? Here’s the idea:

You know how when you drag a categorical attribute (like gender) to an axis, the categories (male and female) appear there? Then, the dots arrange themselves appropriately in that section of the graph, with all the male dots in their own section and the female dots in theirs.

What you might not have thought is: “Each of these sections of the graph is actually a little graph!” They might share an axis (e.g., height or income), but the arrangement of the dots in each section is a mini-graph, filtered (ooh, data move!) by gender, according to the value of that categorical attribute.

This is the idea behind faceting: each mini-graph is a facet of the display, like a facet on the face of a gem.

That’s all ordinary CODAP graph-making, but you might not have realized that you can do additional splits in order to get certain results. The key is to drop categorical attributes, not on axes (which are on the left and the bottom of a graph) but rather at the top or right side of a graph.

Note that any of this splitting by categorical values is an example of grouping.

16.2 [Cautionary] Example

For example, the following graph shows incomes for a few hundred Texans in 2020, broken down by education.

Ha! It sure looks as if you have a better shot at a higher income if you get more education. I wonder if there is a gender difference?

Let’s further group the data by Sex. To do this, we drag Sex to the top of the graph. A yellow region will appear when you’re in the right place, and if you hover there long enough, you’ll see a tool tip. The resulting graph looks like this:

The red arrow shows where we dropped Sex

And then, if we wanted to further group by employment status, we could drag employment status to the right side of the graph:

I show you all this so you can see the technique, but also the danger: this graph contains a whole lot of information, but it’s almost useless because it’s so hard to grasp what’s happening. As I said on the introductory page to this part, it’s too much work for the reader!

16.3 Improving the middle graph

Even the middle graph—the one with education, further split by Sex—is hard to read clearly. The problem is, first, that we have 16 graphs to deal with. But second, you can’t directly compare male and female incomes because you have to refer to two different axes.

Suppose we dragged Sex to the right instead of the top? We get this:

Now we can sort of see that men earn more than women, but what we really want is for the genders to be right next to each other within each education category.

For that, we need to split by Sex first, and then by education:

Now, with a little effort, we can see the gender differences, and how they change with more education.

A big takeaway

This one situation—where you group using two categoricals on the same axis, hierarchically—is one of the most important reasons to use this facet/layout gesture.

The key is to think about which one you drop first, on the “regular” axis. The answer is to drop the “inside” attribute first—but you will never remember that. At least I don’t. So I just resign myself to redoing about half of these graphs. No matter. It’s quick!

The graph is still hard to read, though. If the point is to communicate something about education and persistent gender differences, we should do better. Here are some suggestions for how to improve the message:

  • Reduce the number of education categories (see this section about recoding). You might make it binary: degree/no degree or college/no college.
  • Rescale the axes to suppress large values. This will expand the positions of the means on the plot, making differences easier to see.
  • Abandon the individual points and summarize! (that is, drag Sex and education to the left and make a new column…) Then, plot the medians as bars.
  • Use color to make one attribute clearer.

I did the last two bullets to get this:

Although it’s still a complicated graph, it’s much cleaner, and communicates a message: You make more money the more education you have, but at every level, men earn more than women.

And what data moves did we use? Grouping and summarizing.

If we had recoded the education categories, that would have been calculating (though if my students call it recoding I’m fine!).

And if we had, thoughtfully, removed people who were not in the labor force, that is, used the Employment status information we left out, that would have been filtering.

(Actually, I did use filtering, I realized: I set aside the children with incomes of 9,999,999.)

My point being: in order to take hundreds of raw, multivariate Texans and turn them into a comprehensible display, I needed to alter the dataset in various ways.

16.4 Your turn! Practice with the Texans.

Here’s the data in a live example. Try out these various techniques!