Hey folks,

It turns out that I sent out last week’s newsletter a week earlier than I should have. I had every intention of posting a critique video on Monday and doing a livestream on Wednesday. Alas, I didn’t totally have control of my schedule and neither video got up. Regardless, I’m glad I sent the newsletter out last week because it got me going. Inertia is a powerful thing. If we never start - even if it’s a false start - we will never get going. Rest assured that the videos will come in the following week. The silver lining is that I was able to reach out to the original authors again and they sent me the data I was looking for. In case you missed the last newsletter, you can find it here.

As I post critiques, it never ceases to amaze me how little reviewers critique figures. Many of the figures I critique have the peer-reviews published alongside them. I will often look at those critiques to see what the reviewers had to say about the visualizations. The reviewers may quibble with an interpretation of the data, but they hardly ever say, “You have paired data but have visualized them as though the data were independent, you need a slope plot instead of a box plot”.

Why is this?

In talking about data visualizations with folks my sense is that many people feel that those visualizations ought not be subject to critique because they are a matter of personal style. Whether we use a box and whisker plot or a dynamite plot doesn’t really matter. I prefer one over the other, the author prefers the alternative. That’s fine.

But is it? Sometimes.

When is it a matter of personal preference and style? I would argue that for the most part colors and point shapes are choices that authors are free to express themselves with. Your font choice is largely up to you (unless the journal specifies a specific font). I recently saw a plot with a serifed font. It looked weird, but that’s what the authors wanted. So it goes.

Of course, there are some caveats to this. These caveats help illustrate when our choices shouldn’t be a matter of personal preference and style. We don’t want too many colors or too many shapes because humans are incapable of keeping track of more than 4-7 things in their head at one time. Also, we need to avoid red and green color combinations. We should avoid stereotypes in selecting colors (e.g. blue for men, pink for women). We should avoid using black to denote negative characteristics. We should be aware that we are hardwired to think of certain colors representing positive values (e.g. red) and negative values (e.g., blue).

I sat through a student’s defense the other day when they made a number of questionable choice when it came to color. For example, they had heatmaps where the colors went from yellow to green. That’s an acceptable color scheme. Except that the range of values went from negative to positive. It was really hard to keep track of what color was zero because I’ve been so conditioned to think zero should be a light color like white.

Beyond color and shapes, there are other factors to consider like the experimental design. Are the data paired? Then we need to represent the pairing of the data using a slope plot or ratios of the two conditions. I’m surprised by how much paired data are in the literature and how often people actually depict their data in a paired manner.

I could go on with other examples. And that is largely the point of this newsletter and my YouTube videos.

But, the #1 thing I keep coming back to in the design of figures is that we do a horrible job of making it clear what the point of the figure is. If you’re showing me a heatmap with 1000 rows and 200 columns, what do you want me to see? Is there a better way to show me what you see? Honestly, everyone has a lot of data these days. It’s not impressive to show them off. What is impressive is how you can distill all that data down to a simple story. That is where you can leverage the concepts I discuss to effectively communicate your science.

Next time you’re reading a paper or sitting through a seminar, ask yourself, “What am I supposed to take away from this figure? Why does this figure work? What could make it better?” Then learn from those answers and incorporate those ideas into your own figures. I think you’ll find that’s a lot more rewarding than doing the same tired data visualization tropes that everyone does because they’re what the field expects or because they aren’t thinking very deeply.