WASHINGTON - Support for gay marriage has seen dramatic gains in the United States over the past decade and now has the support of a majority of Americans, a study released Wednesday showed.
The survey by the non-partisan Public Religion Research Institute found that 53 per cent of Americans said they are in favour of gay marriage.
That number reflects a 21-point jump compared to 2003, when 32 per cent of people said they supported same-sex marriage.
The survey reported a sizeable increase in support for gay marriage among many Americans who regularly attend houses of worship, sometimes viewed as being less receptive to gay rights issues.
The poll found majority support for gay marriage among practicing Jewish Americans (83 per cent), white mainline Protestants, (62 per cent); White Catholics (58 per cent) and Hispanic Catholics, (56 per cent).
Hispanic Protestants were split almost down the middle, with 46 per cent saying they were in favour of same-sex marriage and 49 per cent opposed.
Much less support was found among black Protestants (35 per cent) and white evangelical Protestants (27 per cent).
The pollsters said the big increase in support for gay marriage in recent years can be explained in large part by increasing numbers of straight people who say they have a close friend or relative who is gay or lesbian.
The poll found that to be particularly true for Americans in their 20s and early 30s - often referred to as Millenials - who sometimes go so far as to reject a childhood religion they deem to be homophobic.
Sixty-nine per cent of Millenials favour gay marriage, the poll found.
"This new research provides further evidence that negative teachings on this issue have hurt churches' ability to attract and retain young people," said the institute's CEO Robert Jones.
"Nearly one-third of Millennials who left their childhood religion say unfavorable church teachings about or treatment of gay and lesbian people played a significant role in their decision to head for the exit," Jones said.
The survey queried 4,500 respondents between November 12 and December 18, and had a sampling error of plus or minus 1.7 points.