In science, figures often serve as the primary analytical and collaborative space. Communication happens through the figure, not just about it. Often, there isn’t accompanying text to do the explanatory work.
Let's look at this example:
Here's the accompanying text used to caption four of these graphs:
Figure 4: Various distributions of posts, users, and tags.
Thanks for abandoning the reader!
I squinted at the graph and extracted a hand-wavy answer:
The majority of Tumblr posts in the dataset only reached small audiences, typically between 1 and 100 users. Very few posts reached large audiences: only a few dozen reached 10,000 users.
Want to nitpick my interpretation? Great! That’s exactly my point.
Three days later, I am still angry at this graph. It’s forcing the reader (ie. me) to do all the labour.
Number of posts versus number of users per post. What does this metric even mean? Uh… my brain stalled because it had to convert this phrase into something meaningful. Okay, this is a distribution of reach, but not one I would naturally think about. Oooh boy. How many posts had exactly 37 users?
Then the log scale crushes the only part that matters. Most posts sit between 1 and 100 users, but you can barely see it. Everything collapses into a red smear, so you can’t tell where the data actually clusters. And nothing is highlighted, so what’s important here?
So, the reader ends up squinting, translating axes, mentally re-binning, and then writing the outcome statement themselves. This is why people launch themselves into graph avoidance.
