Ambinder points to a sort of surprising anomaly in a recent poll on gays in the military: substantially more people support letting "Gays and lesbians" serve in the military than "homosexuals". When I mentioned this to someone this morning, he cocked his head to the side and stood stock still for about five minutes, trying to wrap his brain around this.
But perhaps it's not that surprising, given everything else we know about framing effects. "Gays and lesbians" sound like people. "Homosexuals" sound like a medical condition. I'd bet you could get similar differences if you polled increased disability benefits for "schizophrenics" versus "people with a serious mental illness". We like helping people. But in general, we dislike helping groups.
Hell, I bet tax increases on "the rich" poll better than tax increases on people who make more than $250,000 a year--and not only because the poll-ees who live in Manhattan tend to think of a quarter of a million dollars as scraping by on the edge of poverty.
The real question is: which is a better approximation of the public's "true" opinion? Obviously, if you support gay marriage, the temptation is to make the case for the more "people-like" option, but I could make a case for the other side. Maybe a better question is whether there is any "true" opinion . . . or whether, as with Schroedinger's cat, the answer only comes into existance at the moment you actually ask the question.