NYU Political Science Professor Patrick Egan has studied 10 years worth of public opinion polling conducted during election campaigns involving same-sex marriage ballot questions, and found that such polls (even those conducted just before election day) consistently overestimate voter support for gay marriage. Comparing pre-election polls with the election results, Professor Egan's study concluded that the level of opposition to gay marriage reported in polls tended to be accurate, but that there was typically a 7 point shortfall between expressed support for equal marriage rights and election day results.
Takeaway: Because pollsters create a bottom line prediction by allocating the "undecided's" in the same proportion as the "decided's," polls have consistently overestimated the number of voters who support marriage equality, by an average of 3 per cent.
Put another way, either a chunk of voters tell pollsters they will support marriage rights but then don't, or essentially all of the undecided's vote against marriage equality. The study's significance lies in the repetition of this finding - across 10 years of rapidly changing public opinion and 28 different states from Maine to Alaska and Hawaii. Most of the polls analyzed involved proposals allowing amendment of the state's constitution to ban same-sex marriage, now adopted by voters in 30 states.
Professor Egan rejects what seems the most obvious explanation for the gap: the "Bradley effect," in which voters hesitate to admit that they intend to vote in a biased way and profess an intention to vote otherwise, but then do not. (The name references former LA Mayor Tom Bradley, who "won" the exit polls in his 1982 CA gubernatorial race but lost the actual election; Bradley, now deceased, was African-American.) Egan found no difference in polls about gay marriage between those conducted by a human interviewer and those conducted by automated systems, thus discounting the Bradley effect as the cause of the differential.
Professor Egan told a press conference hosted by the Haas Jr. Foundation, which funded the study, that it was impossible to know without further investigation why the discrepancy occurred. He suggested that it might be because opinion polls during campaigns screen for "likely voters," and that those screens might operate in a way to filter out more voters opposing same-sex marriage than voters supporting it. Or, it might be that for this issue, voters who say they are undecided almost all vote the same way in the end: to enact bans on gay marriage. The latter explanation would be odd, he said, because if true, data show that the phenomenon that "undecided = opposition to marriage equality" does not fluctuate throughout the course of the campaign.
Another finding drawing much attention was that neither side's advertising appears to have a significant effect on voting patterns. Professor Egan pointed out that this finding does not indicate what would have happened in the various campaigns had one side done no advertising. However, it does cast big doubt on the arguments among lgbt advocates about which kinds of ads (showing gay families, not showing gay families, etc.) have greater impact during a campaign. Apparently, neither the content nor even the ad campaigns as a whole, for either side, ended up producing any significant effect on changing voter attitudes.
Equality California ED Geoff Kors summed up the lesson for future elections, specifically in the context of repealing Prop 8: "Before we go back to the voters, we need to see solid majority support. Having 49% or 50% support is not enough."
Professor Egan's database included 167 surveys conducted in reference to 32 ballot measures - 30 involving marriage and two related to domestic partnership laws. The database included the No on 8 campaign's internal polls from the 2008 election.
For the first time in many years, there will be no ballot initiatives regarding same-sex marriage or domestic partnership in this year's election.