Additional Info: I designed a Python workflow to perform OCR on every xkcd comic, feed that text into a large language model, and ask the model whether this comic was about the category named in the title.
Report an error
xkcd comics published about statistics correlates with...
Variable | Correlation | Years | Has img? |
The number of concierges in Ohio | r=0.92 | 16yrs | Yes! |
Votes for Republican Senators in West Virginia | r=0.92 | 6yrs | No |
The number of construction laborers in Mississippi | r=0.89 | 16yrs | No |
Total views on LockPickingLawyer YouTube videos | r=0.88 | 9yrs | No |
Votes for Republican Senators in Delaware | r=0.88 | 6yrs | No |
The number of detectives and criminal investigators in South Carolina | r=0.86 | 16yrs | No |
Total length of OverSimplified YouTube videos | r=0.86 | 7yrs | No |
Liquefied petroleum gas used in Bahrain | r=0.78 | 15yrs | Yes! |
US household spending on fresh vegetables | r=0.73 | 16yrs | No |
Shark attacks in the United States | r=0.58 | 16yrs | No |
xkcd comics published about statistics also correlates with...
<< Back to discover a correlation
You caught me! While it would be intuitive to sort only by "correlation," I have a big, weird database. If I sort only by correlation, often all the top results are from some one or two very large datasets (like the weather or labor statistics), and it overwhelms the page.
I can't show you *all* the correlations, because my database would get too large and this page would take a very long time to load. Instead I opt to show you a subset, and I sort them by a magic system score. It starts with the correlation, but penalizes variables that repeat from the same dataset. (It also gives a bonus to variables I happen to find interesting.)