The most frightening thing isn't that 14% of Americans think sound travels faster than light. The most frightening thing is that if you flipped a coin to decide whether to believe that statement or its converse, you'd land on 'speed: sound > light' half the time, in which case zero evidence impinges on the decision... which means you could just as easily have randomly believed 'speed: light > sound'.
Thus, at least 28% of Americans have no clue.
It gets worse when, rather than being a binary choice, there are several "choices" of alternative beliefs.
I'm not sure this quite follows, as it isn't clear that the 14% were guessing randomly... they might have been operating from some systematically wrong model. But I agree that it's likely.
To be fair, I also often wonder whether the people who take these tests are just picking random answers because they aren't at all invested in the test. Back when I was taking neuropsych evaluations after my stroke, there were a number of questions intended to detect malingering; I would like to believe that tests like those cited here have similar "checksum" questions built in and that reported results would take that into account, but in fact I don't.
Encountered at: https://whyevolutionistrue.wordpress.com/2011/03/20/click-and-weep/scientific-literacy/