I think that there are better analyses of calibration which could be done than the ones that are posted here.

For example, I think it's better to combine all 10 questions into a single graph rather than looking at each one separately.

The pattern of overconfidence on hard questions and underconfidence on easy questions is actually what you'd expect to find, even if people are well-calibrated. One thing that makes a question easy is if the obvious guess is the correct answer (like a question about Confederate Civil War generals where the correct answer is Robert E. Lee). On those sorts of easy questions, a bunch of people who made the obvious guess with moderate confidence will turn out to be correct, and they'll look underconfident. Whereas on questions where the correct answer is more obscure or counterintuitive, those guesses will be wrong and they'll look overconfident.

I'll see what other analyses I can do with the data. First I'll need to make some not-entirely-straightforward decisions about what to do with misspellings and other interesting responses. This may be fairly important, since those account for a decent chunk of all responses. For example, here are the most common responses to question 8 (in descending order of frequency):

Mitochondria
mitochondria
Mitochondrion
mitochondrion
Nucleus
Mitocondria
mitochondrium
nucleus
mitocondria
Mitochondrium
Mithocondria
Mitochondira
Ribosome