Several of the previous comments here mention the methodological issue of how much these data really represent the population of interest. A FAQ comment posted to Hacker News before[1] points out that voluntary response data in general are often very unreliable for generalizing to nonresponders. We can't be sure that the people who bother to respond to online quiz questions or poll questions (including here on Hacker News) have much resemblance to the people who do not bother to respond.