Visualizing Correspondence Analysis: Heatmaps or Scatter Plots?
Correspondence analysis is unrivaled when it comes to helping researchers identify meaningful relationships in data. It uncovers hidden patterns and connections between data, even when you don't know what you are looking for. These insights are only as valuable as the way they are visualized.
Correspondence analysis is all about summarizing patterns in a table of data as a visualization. But how exactly should you visualize this data?
Heatmaps are an obvious choice. They're visually striking, easily digestible, and great with large sets of data. But when it comes to visualizing correspondence analysis, there's a reason why they are not the industry standard.
Visualizing correspondence analysis
Correspondence analysis is an incredibly powerful data analysis technique. It's how researchers analyze the relationship between two or more categorical values.
This makes it great for connecting brands with customer opinions or products with preferences. If you want to learn more you can read Displayr's explainer on the underlying math of correspondence analysis.
Although correspondence analysis is considered a statistical method, it is all about how you visualize the findings. Consider the below. It's a correspondence analysis about the personality associations of different beverages.
It shows us almost nothing, unless you really know what you're looking for.
Heatmaps are great, except when...
If we were to visualize the table above as a heatmap, we might encounter some problems. Heatmaps can display an optical illusion known as the checker shadow illusion, in which we can notice sharp contrasts in shading when the two colors are adjacent on a visualization, but not so much when they are non-adjacent. The video below illustrates the illusion.
When two dramatically different values are placed next to each other, the contrast jumps out — your eyes are drawn to the boundary. But when two interesting or important values are separated by several rows or columns, you might miss them entirely. As you can imagine, this can be problematic if you're working with correspondence analysis.
It's important to note that heatmaps are still a powerful way to visualize data. They look great, display information in a way that is easy for our brains to understand, and can show categorical labels or numeric values (although correspondence analysis only handles categorical). This is why they are such an effective way to make big tables of data easier to read.
Recommended by LinkedIn
Scatter plots: the gold standard for correspondence analysis
A scatter plot is a graph that shows the values of two different variables as points, with the data for each point represented by its horizontal and vertical position.
Different values are typically represented with labelled dots, however, sometimes replacing labels with images (e.g., logos) can improve the visualization of a correspondence analysis, as it makes them more attractive and easier to digest.
Scatter plots shine when it comes to visualizing complex tables. Take the following example. The table shows 28 brands and 15 different brand attributes. To gain any sort of insight from the table itself is almost impossible.
The scatterplot shows how certain brands align with specific personalities. We can see Calvin Klein, American Express, Apple, and Lexus are all Upperclass, while Nike, Reebok, Levi's, and Michelin are tough.
By plotting each variable as a point on a graph, you can get a more accurate understanding of the relationships. For example, if two brands appear close together on the chart, it means they have similar profiles. If a brand is located near a particular trait or attribute, it suggests a strong association. And anything near the center? That usually means it's less distinct — not strongly tied to any one trait or group.
The verdict
When it comes to visualizing correspondence analysis, starting with a scatter plot is usually the safest option. It provides a clear, structural view of relationships — the very essence of what correspondence analysis is designed to uncover.
That said, this doesn't mean you should completely avoid heatmaps or any other visualization technique. Each chart has its strengths, and choosing the right one depends on the data story you are trying to tell.
Some other visualizations you can try include:
Want to learn more about making the most of correspondence analysis? Check out Displayr's hands-on guide to correspondence analysis.