The results don’t reflect that people think 1 out of 4 are homosexual. The results reflect that Americans don’t understand simple mathematic principles.
People generally don’t think in terms of probabilities—this poll also provides categories when the underlying metric is a continuous variable from 0 to 100 percent. If I ask a question that I have information suggests the most probable correct value is about 2 percent but then provide a set of categories that nearly all are higher, you would find people generally would pick a category near the middle of the set. This is old stuff in psychology research and it is a good illustration of the difference between polling and scientific measurement.
A key clue is that the more education the respondent has, the lower the estimated percentage. This is probably a better test of a person’s knowledge, than of the issue it claims to measure.