Just Post

Given that the "estimated proportion" range is only between 20-76%, regardless of the question, this seems more indicative of a poor understanding of statistics than an over/under estimation of specific demographics, especially since a lot of contradictory demographics are way overestimated.

For example, I am significantly more likely to believe that Americans suck at percentages than that they believe nearly ever single person in the country is either Muslim, Jewish, or Catholic (these three "estimated proportions" add up to 98%).

Side note: interestingly all religious categories listed add up to 189%, but there is some overlap depending on definition (e.g. some people might argue that "Catholics are Christians" or "Jewish is an ethnicity so you can be Jewish and atheist"). Thus I picked the three that most people would agree are extremely unlikely to overlap, which coincidentally added up to nearly 100%.

EDIT: I would like to see this survey redone with the same questions, but with the addition of a few questions for things that are widely known to be extremely uncommon, bordering on non-existent, such as "percentage of people with only one ear" or "percentage of people with more than 12 siblings," and some questions for things that are widely known to be extremely common, bordering on universal, such as "percentage of people that have electricity at home" or "percentage of adults who own a phone." If even these questions result in answers grossly over/underestimating the percentage, what we have is actually an aversion to providing very small or very large estimates. (It is already known that people easily overestimate the frequency of things that are unusual especially if they can easily think of an example, such as overestimating the number of redheads because you had a classmate with red hair, or even because you can think of a celebrity with red hair).