If Same-Sex Marriage Is so Popular, Why Does It Always Lose at the Ballot Box? (Includes state-level data on support and legislation)

With the continuing debate regarding the electoral implications of Obama’s announcement regarding his support for gay marriage, we are very pleased to welcome the following guest post from Gregory B. Lewis of the Andrew Young School of Policy Studies at Georgia State University:

Since all 31 states that have voted on constitutional amendments to ban same-sex marriage (SSM) have passed them, typically by overwhelming popular votes, should we be skeptical that half of Americans really support same-sex marriage?  Probably not.  Most bans passed when opposition to SSM was much stronger, and SSM opponents have targeted constitutional amendments for votes in states where support for SSM is weakest.

Opposition to SSM was quite strong and reasonably stable until 2004.  Since 2004, the rise in support has been remarkable.  My estimate is 16 percentage points.  Nate Silver estimates perhaps two or three percentage points a year and, according to a leaked memo, Republican pollster Jan van Lohuizen finds support rising one point a year until 2009 and 5 points a year since.  Seventeen states passed constitutional amendments by the end of 2004, and 27 did so by 2006.  Even in 2008, when next three states passed amendments, support for SSM nationally was probably 8+ percentage points lower than it is today.

Opposition to SSM varies widely by state.  Seong Soo Oh and I concluded that support was 30 points higher in Massachusetts than in Mississippi in 2006.   Jeffrey Lax and Justin Phillips found a 40 point split between Massachusetts and Utah in 2009.  My most current estimates find nearly a 50 point division between Massachusetts and Mississippi.

Constitutional amendments reach voters mostly in states where support for SSM is weakest: 23 of the 29 states with the least support have approved constitutional bans.  In contrast, 18 of 20 states with the highest popular support for SSM legally recognize same-sex couples in some way.  North Carolina has the 12th lowest support for SSM, making its passage of Amendment One last week less surprising than its waiting so long to do so.

Popular votes against constitutional bans are generally close to my estimates, within three percentage points in the four votes since 2008.  In 28 of 31 states, I under-predict the percentages voting against.  (Nevada opposition was two points lower than expected, and I blame my atrocious Alaska prediction on a very small sample and a very early vote.)

In 2011, majorities supported SSM in 15 states and the District of Columbia, and another five states may reach majorities this year.  If constitutional bans came up for votes in California, Colorado, and Oregon this year, they would probably fail.  Nevada, Hawaii, Wisconsin, and Arizona would be toss-ups.  In Minnesota, which will vote on a ban this fall, the estimated 49.9% support for SSM makes the vote too close to call.

Info on data and methods after the break:


I use data on 124,032 respondents to 102 national surveys conducted by 12 polling firms between August 1992 and December 2011, most of which I obtained from the Roper Center for Public Opinion Research.  The surveys used 22 different question wordings.  My dependent variable is coded 1 for those who supported SSM and 0 for those who gave any other response (including “Don’t Know” and “Civil Unions” for those asked whether they favored marriage, civil unions, or no legal recognition).  I ran a logit model with 21 dummy variables for question asked, 50 dummy variables for state, and 15 dummy variables for survey year.  I created an artificial data set with one line for each state for each year, coding the question dummies as if all respondents had answered the most commonly asked question:  “Do you think marriages between same-sex couples should or should not be recognized by the law as valid, with the same rights as traditional marriages?”  I then used the predict command in Stata to estimate the percentage supporting SSM in each state in each year.

To estimate recent time trends, I re-ran the analysis using OLS regression, keeping the dummy variables for years up through 2004 and adding 51 linear time variables, one for each state, for the period since 2004.  As the vast majority of predicted probabilities are between 20% and 80%, the “linear” portion of the logit response curve, OLS generates virtually the same estimates as logit and the interpretation is much easier.

[Cross-posted at The Monkey Cage]

Joshua Tucker

Joshua Tucker is a Professor of Politics at New York University.