Our Maine poll tomorrow is going to show a much rosier picture for opponents of same sex marriage than two other polls released recently by Pan Atlantic SMS and Democracy Corps, which showed the referendum failing by an average of 51/42.
It is of course always possible that we're wrong. But I think it's more likely that this is one issue where automated polls might have an advantage over live interviewers. There is probably some segment of the population that knows opposing same sex marriage is wrong and not intellectually defensible but still does. Those are the kinds of folks who might tell an interviewer what they believe they should think rather than what they do think.
Perhaps someone with more time can do a more thorough look at how polling on gay marriage and affirmative action bans reflected the actual results, but on Proposition 8 in California last year three IVR polls from SurveyUSA showed it passing by an average of 47/46 while the Field Poll and Public Policy Institute of California, both using live interviewers, showed it failing by an average of 51/44. It's a limited sample size but there was an eight point difference on the margin between live interview and automated polls. And of course the ultimate result, passage by a margin of 52-48, was not something any pollster showed in the last couple weeks of the campaign.
Because the referendum is so close we're probably going to sneak in another poll of it between now and November 3rd.