Hey folks,

The more I peruse the literature, the more I see that researchers need help designing figures to help tell their stories. I don’t just mean the mechanics of creating a figure in R, Python, Prism, or Excel. Rather, if someone had a box of dry erase markers of various colors and they had to give a talk without any slides, what would they draw to tell their story? I don’t mean to trivialize the difficulties. It’s hard! There are many figures I’ve published that I wish I could have a “do over” on. If you’re interested in working with me on an individual or group level to improve your data visualizations, let me know. I provide free 30-minute exploratory meetings to discuss how we might work together to design the figures for your next paper, talk, or poster. You can sign up by clicking the button below!

Pop quiz! Do you know how many figures Watson and Crick included in their 1953 article, “Molecular Structure of Nucleic Acids: A Structure for Deoxyribose Nucleic Acid”, which was published in Nature? 1. It’s actually a diagram. There really no data.​

Let’s fast-forward 73 years. I have found what I think is the figure with the most lettered panels I’ve seen. It was recently published in the article, “A brain reward circuit inhibited by next-generation weight-loss drugs in mice”, which was also published in Nature.

This figure has 26 lettered panels - a through z. Figures 2 and 4 “only” have 25 lettered panels. Figure 5 has 22 and Figure 3 has 14 lettered panels. Some of these panels have facets, which would push the number of panels even higher. Among the 5 main figures, this paper has 112 of what Watson and Crick and most normal people would consider figures.

In this case, I’m sure that many of the panels could be combined into a single plot to actually tell a better story. Regardless, the phenomenon of bloated numbers of figures in papers is common in the biomedical sciences. Why is this happening? I don’t know. I’m not sure that I’ve actually ever had more than 4 or 5 lettered panels in a figure for a paper that has come from my research group. Here’s what I suspect authors with these large number of panels would have to say

In 2026, it requires a lot of data to publish a paper in Nature. Perhaps. But I think this is BS. If you look back a couple of weeks to a figure I posted in this newsletter you’ll see a pie chart where the only category occupied 100% of the circle. I’ve also discussed whether a plot is worth 1,000 words or even 100. On Monday I will post a critique video showing how the 26 panels I included above is certainly inflated. Although large and complex datasets are increasingly common and perhaps “required” to publish in Nature, Cell, and Science, the number of panels is certainly inflated.

To overcome fears of the “reproducibility crisis”, we need to show all the data. I’ve heard this as an argument for why we also need P-values with 10^-4 or 10^-17 precision even if we’re comparing mutant and wild type mice with three animals per group. I don’t buy these arguments. A pie chart with one number, bar plots with two points, artificially separating treatments into separate plots doesn’t make the analysis more transparent or reliable. If that’s the concern, then why don’t more people make their data available? Why don’t people publish the actual code they used to generate their figures? Why do people insist on using Illustrator to assemble these behemoths? Those are true threats to reproducibility.

Publishing in Nature is hyper-competitive and we need to overwhelm reviewers and editors with data to convince them of the work’s impact. This might be too cynical, but I think it’s the most honest answer. I forget who told me this, but there’s a difference between data and results. It’s a results section, not a data section. Synthesize the data for your audience. Tell me what you expect me to see in the data. The figure above has a descriptive title, “Validation of small-molecule GLP1R agonist-responsive mouse model”. The caption has descriptive titles that border on jibberish, “f–h, Glucose (dextrose; Dex) tolerance test. Blood glucose levels on liraglutide (f; WT n = 8, Glp1rS33W n = 6), danuglipron (g; WT n = 9, Glp1rS33W n = 6) or orforglipron (Orfo)”. They aren’t showing the results, they’re showing data.

Considering there’s at least another 120 figures in the supplement, that’s more than 200 total figures in the paper. If a reviewer is doing their due diligence, it would likely take them an average of one minute per panel or 200 minutes. Perhaps one minute is too much, let’s call it a total of 3 hours. Just for the figures. I think it would take them another hour or two to review the text. I’d love to meet the person that regularly spends half their day reviewing a manuscript and is willing to entertain multiple rounds of review (as a rule, I don’t review for any CNS journals).

Overwhelming reviewers and editors with data is another threat to the reproducibility crisis. When gatekeepers get rushed, they make bad decisions. They don’t notice things that they should have. I spend several hours prepping and recording my critique videos. Sometimes, I don’t actually notice a problem with a visualization until I’ve tried to recreate the visual because finally noticed it after 5 hours or because I’m working with the actual data. No one is going over all the figures in a paper like this.

What are the solutions? First, smaller papers. Perhaps they aren’t as “impactful” (whatever that means), but they are more trustworthy and easier to digest. Second, divorce “impact” from the journal. Only assess impact 5 years after a paper has been published. Finally, focus on results and not data.

What do you think? Can you think of other reasons why we have seen an escalation in the number of panels in scientific papers? Have you seen more than 26 lettered panels in a figure? Reply to this email and let me know!

​

​