research paper about Gay Marriage should not be prohibited.

Religion and Public Opinion about

Same-Sex Marriagen

Laura R. Olson, Clemson University

Wendy Cadge, Bowdoin College and Harvard University James T. Harrison, Bowdoin College

Objective. The goal of this article is to analyze the relationship between religion, measured in terms of religious affiliation and religiosity, and public opinion about same-sex marriage, civil unions, and a federal constitutional amendment that would prohibit gay marriage. Methods. We use logistic regression with calculated standardized coefficients to analyze data from a nationally representative survey of 1,610 respondents conducted in March–April 2004. Results. Religious variables perform better than demographic measures in models of attitudes about same-sex unions. Non-Protestants are much more likely to support same-sex unions than are Protestants, and individuals with conservative attitudes toward morality and secularism and (to a lesser extent) those who participate actively in religious life are more likely to oppose such unions. On the whole, religious variables play a weaker role in predicting support for a constitutional amendment to prevent gay marriage than they do in predicting attitudes toward same-sex unions. Conclusions. Religious variables play powerful roles in structuring attitudes about same-sex unions. Moreover, homosexuality appears to be a major component of the ‘‘moral values’’ discourse that is currently so popular in American politics.

Exit polls conducted on Election Day 2004 (and much debated since then) show that 22 percent of the American electorate claimed to be motivated by moral values when they turned out to vote in the presidential election between George W. Bush and John Kerry (Edison Media Research and Mitofsky International, 2004). Gay marriage was one of the most significant moral issues lurking below these exit poll results. Recent analyses show that the issue had a significant influence on individual voters and state vote totals in several regions of the United States (Lewis, 2005).

nDirect correspondence or requests for data and coding information to Laura R. Olson, Department of Political Science, Clemson University, Clemson, SC 29634-1354 hlaurao@ clemson.edui. We thank Robert Wuthnow for providing us with the data for this project. Additional thanks are due to two anonymous reviewers, Jason Barabas, Paul Djupe, John Green, Robert Lineberry, Nina Paynter, Joe Stewart, and Adam Warber for their helpful advice and input. An earlier version of this article was presented at the Eastern Sociological Society meetings in March 2005.

SOCIAL SCIENCE QUARTERLY, Volume 87, Number 2, June 2006

r2006 Southwestern Social Science Association

Public debates and related court battles about gay marriage began in the United States more than 10 years before the 2004 presidential election (Andersen, 2005; D’Emilio and Freedman, 1988; Mello, 2004; Strasser, 1999). The issue took on renewed national visibility recently in response to Goodridge v. Department of Public Health (2003), which legalized same-sex marriage in Massachusetts. People of faith, especially evangelical Protestants (who have long been a prominent force in anti-gay rights efforts in the United States), mobilized in response to the Goodridge decision, successfully leading efforts to amend numerous state constitutions to prohibit same-sex marriage (Green, 2000; Herman, 2000; Soule, 2004). Led largely by religious conservatives, voters in 11 states (Arkansas, Georgia, Kentucky, Michigan, Mississippi, Montana, North Dakota, Ohio, Oklahoma, Oregon, and Utah) approved related amendments in the November 2004 elections, and legislation and litigation in several other states is pending (see, e.g., Damore, Jelen, and Bowers, forthcoming). At the federal level, resolutions were introduced in Congress following the Goodridge decision to amend the U.S. Constitution to restrict marriage to heterosexual couples. The U.S. Senate rejected the Federal Marriage Amendment in July 2004, though some activists continued to push for such an amendment in 2005 (Easton, 2005; Liu and Macedo, 2005).

A large scholarly literature describes the relationship between religion and public opinion about homosexuality. Absent from this literature, however, is specific analysis of how religion influences public opinion about same-sex marriage. Such analyses are important not only because of the issue’s current political resilience but because of the insights these debates provide into broader theoretical concerns about the possible reemergence of the culture war (or at least its rhetoric), the emerging ‘‘God gap’’ in American politics, and (indirectly) the processes of social movement activism. We rely on unique survey data (N51,610) collected in March–April 2004 to examine how religion (as measured by both religious affiliation and religiosity) influences attitudes about gay marriage, as well as attitudes about civil unions and a Federal Marriage Amendment that would restrict marriage to heterosexual couples. This article marks the first systematic analysis of religion and public opinion about same-sex marriage in the post-Goodridge United States and allows considerable nuance in interpretation because of the multiple ways religion and public opinion about same-sex unions are measured.

A good deal of research has been conducted about the relationship between religion and public opinion about homosexuality generally, with particular attention to changes in public opinion and the significance of demographic and religious predictors over time. Beginning in the 1970s, surveys assessed Americans’ attitudes about the morality of homosexuality and their attitudes about restricting the civil rights and civil liberties of gay people (see, e.g., DeBoer, 1978; Levitt and Klassen, 1974). Jeni Loftus (2001) showed, based on data from the General Social Survey, that Americans’ attitudes about the morality of homosexuality became more liberal between 1973 and 1976, more conservative between 1976 and 1990, and more liberal again between 1990 and 2001 as a result of demographic changes and cultural shifts. Since the 1970s, Americans on the whole also became less willing to restrict the civil rights or liberties of gays and lesbians.

Recent polls indicate that demographic factors such as education, gender, and age have significant influences on public opinion about homosexuality, as does the degree of personal contact individuals have with gay men and lesbians, and attitudes toward traditional morality (Brewer, 2003; Davis, 1992; Ellison and Musick, 1993; Finlay and Walther, 2003; Gibson and Tedin, 1988; Glenn and Weaver, 1979; Herek, 2002; Herek and Capitanio, 1996; Herek and Glunt, 1993; Kerns and Fine, 1994; Kite and Whitley, 1996; Loftus, 2001). Religion, as measured by individuals’ religious affiliations, behaviors, and beliefs, also has a clear and consistent influence on their opinions about homosexuality. Religious affiliation has an especially strong impact: Jews, liberal Protestants, and people who are religiously unaffiliated have the most liberal attitudes, in part because many of their religious traditions have not systematically condemned homosexual behaviors in recent years. Catholics and moderate Protestants tend to espouse moderate but generally tolerant attitudes. Evangelical Protestants have the most conservative attitudes, reflecting their theological beliefs and official denominational and congregational positions on homosexuality (Cochran and Beeghley, 1991; Cotten-Huston and Waite, 2000; Finlay and Walther, 2003; Fisher et al., 1994; Glenn and Weaver, 1979; Herek and Glunt, 1993; Irwin and Thompson, 1977; Kirkpatrick, 1993; Roof and McKinney, 1987). Religiosity, as measured by frequency of attendance at religious services, is also a significant predictor of individuals’ opinions about homosexuality. People who attend services frequently have more conservative attitudes, at least in part because many of them are evangelical Protestants (Beatty and Walter, 1984; Cochran and Beeghley, 1991; Fisher et al., 1994; Herek, 1984; Herek and Glunt, 1993). Within religious organizations, researchers have also tried to understand the role that reference groups, friendship networks, and other means of group support have on opinion about homosexuality. These studies suggest that people whose social networks are deeply tied to a religious congregation tend to be less accepting of homosexuality. The more close friends people have in their congregations, the more their outlook on life appears generally to be structured by the prevailing sentiment of these friendship networks, which tend not to foster tolerance for social difference (Petersen and Donnenwerth, 1998).

Many of the analyses that assess the relationship between religion and public opinion about homosexuality are based on small surveys, restricted either by geography or by the types of people included (often college students). Nationally representative surveys that allow detailed analysis of the relationship between religion and homosexuality are often limited because most public opinion data includes little information about religion. The Gallup Organization, for example, does not regularly subdivide Protestants into the evangelical and mainline groups needed for careful analyses. Although existing analyses point to suggestive patterns between education, gender, age, contacts with gay or lesbian people, and a range of religious measures, they also do not allow for specific investigation of how these factors influence public opinion about gay marriage, civil unions, or a federal constitutional amendment prohibiting same-sex marriages—three issues that are central to recent public debates about gay marriage. Other than the data presented here, the only other national survey (conducted by the Pew Forum on Religion and Public Life in October 2003) to collect information about these issues pointed to strong religious undercurrents in the shape of public opinion about gay marriage and civil unions, but did not assess these relationships in a multivariate context (Pew Forum on Religion and Public Life, 2004b).

To examine the relationship between religion and attitudes toward samesex marriage, we examine data collected in a telephone survey designed and administered by Greenberg Quinlan Rosner Research, Inc. The survey, which reached 1,610 adults aged 18 years or older, was conducted March 16–April 4, 2004, about one month after media attention focused on same-sex marriages being performed in the City of San Francisco. The PBS television series Religion & Ethics Newsweekly commissioned the survey.1

Random-digit dialing was used to obtain telephone numbers, allowing access to all listed and unlisted phones. The sample was stratified by state. The nationally representative sample of 900 was supplemented with oversamples of 401 white evangelical Protestant respondents, 160 AfricanAmerican respondents, and 149 Hispanic respondents. The data were weighted by gender, age, race, region, and religion. The multivariate analyses presented below are conducted with the weighted data.

The Greenberg Quinlan Rosner survey asked respondents to indicate their attitudes about three distinct aspects of the same-sex marriage debate. First, respondents were asked: ‘‘Do you strongly favor, favor, oppose, or strongly oppose allowing gays and lesbians to marry legally?’’ Second, they were asked: ‘‘Would you strongly favor, favor, oppose, or strongly oppose a law that would allow homosexual couples to legally form civil unions, giving them some of the rights of married couples?’’ Third, only those respondents who stated that they opposed or strongly opposed gay marriage were asked: ‘‘Should the U.S. Constitution be amended to ban gay marriage, or is it enough to prohibit gay marriage by law without changing the Constitution?’’ As Table 1 shows, 60.9 percent of the sample said they either oppose or strongly oppose gay marriage; an even larger proportion oppose or strongly oppose civil unions (72.6 percent). These opponents of same-sex unions are split on the question of how to prohibit gay marriage under law, however; only two in five favor a Federal Marriage Amendment as opposed to statutory prohibitions against gay marriage.

Because we are investigating the extent to which religion shapes public opinion on gay marriage, our principal independent variables tap two dimensions of religious life: religious affiliation and religiosity. Scholars have measured the role of religion in social and political life using religious affiliation; indeed, affiliation is an important component of individuals’ religious outlooks. It is crucial to employ a careful and thorough measure of religious affiliation. The simplistic ‘‘Protestant, Catholic, Jewish’’ framework lacks depth in measurement. Thus we operationalize religious affiliation using the RELTRAD procedure developed by Brian Steensland and colleagues (2000), which classifies individuals into seven theologically distinct and relevant categories: African-American Protestant, evangelical Protestant, mainline Protestant, Roman Catholic, Jewish, other religion, and unaffiliated. We expect to find evangelical Protestants and African-American Protestants to be especially opposed to same-sex unions because their religious traditions consistently teach that the practice of homosexuality is sinful. Table 1 indicates that evangelicals comprise more than one-third of the survey sample. They are therefore overrepresented (as was the intention of the pollster), whereas mainline Protestants and Catholics are somewhat underrepresented (cf. Kohut et al., 2000).

The political relevance of religion cannot be measured by religious tradition alone, however, so we also include measures of religiosity in our analyses. Increasingly, we are hearing in the popular press (Eastland, 2004; Wallis, 2004) and from scholars (Green, 2004; Green et al., 2004; Kohut

TABLE1

Descriptive Statistics

Total N 5 1,610.

SOURCE: Greenberg Quinlan Rosner poll for Religion & Ethics Newsweekly (2004).

et al., 2000; Layman, 2001) that religious affiliation is less important as a predictor of political attitudes and affiliations than the extent to which individuals are committed to and engrossed in religious life. The key measure of religiosity is frequency of attendance at worship services, and we include this variable along with other measures of religious activity in a ‘‘religious activity’’ index (see Guth et al., 1997). In addition to worship attendance, the index includes measures of the frequency with which the respondent: attends informal religious or prayer groups; talks about religion informally with friends; prays before meals; reads religious books, newspapers, or magazines; spends time in prayer or meditation; reads holy scriptures; and watches or listens to religious broadcasting.2 We expect to show that involvement in religious activity increases opposition to same-sex unions because voices of organized religion do not frequently speak out in support of gay couples (but see Cadge, 2002). The more involved one is in organized religion, the more likely he or she might be to hear messages denouncing same-sex unions from the pulpit, in congregational forums and adult education classes, and in the course of informal discussion.

In preliminary bivariate analyses (data not reported), we found that respondents across religious traditions who are involved and invested in religious life have markedly different attitudes toward same-sex unions than their counterparts who do not prioritize religious activities. These results therefore support the emerging ‘‘God gap’’ pattern in American public opinion and voting behavior (Green, 2004; Green et al., 2004; Kohut et al., 2000; Layman, 2001), which demonstrates that Americans who are significantly involved in religious life are more politically conservative than their secular counterparts.

Finally, we include a more subtle measure of the extent to which the respondent’s life revolves around a religious congregation in the form of an item that asks how many of the respondent’s closest friends belong to his or her congregation. This measure taps the depth and strength of one’s informal religious social networks (see Gilbert, 1993; Djupe and Gilbert, 2002). The more close personal friends one has in his or her congregation, the closer one’s psychological and emotional ties are likely to be to the congregation—and therefore the more we might expect his or her participation in congregational life to shape and affect political outlook. In particular, the more congregational friends respondents have, the more likely they might be to oppose same-sex unions because of their personal emo-

2 Except for worship attendance, each of these items has a response set of ‘‘every day,’’ ‘‘once or twice a week,’’ ‘‘once or twice a month,’’ ‘‘once or twice a year,’’ and ‘‘never.’’ We created the index by first assigning each of these responses a score of 4, 3, 2, 1, and 0, respectively, for each item and adding together the scores for each of the items. We then added 0–4 points for the respondent’s answer to the question ‘‘How often do you attend religious services?’’ for which there was a slightly different response set: 05never/hardly ever; 1 5several times a year; 25once or twice a month; 35 once a week; 45more than once a week. Scores on the resulting index range from 0 to 32. The mean score is 17.4 and the standard deviation is 8.5. The index is highly reliable: Cronbach’s alpha 50.89.

tional and psychological investment in the congregation and its teachings (Petersen and Donnenwerth, 1998). Since most religious traditions oppose homosexuality, the messages one hears in one’s congregation about gay couples are not likely to be fully supportive. Table 1 indicates that two in every five respondents (39.7 percent) report that at least half their closest friends come from their congregation.

We note that some of these independent variables are significantly associated with one another (although bivariate analysis shows that no two independent variables are extremely highly correlated; data not reported). Catholics lead other Christian groups in terms of number of close congregational friends (26.7 percent say that most or all of their close friends are from their congregation, compared to 23.9 percent of white evangelical Protestants, 20.0 percent of African-American Protestants, and 17.6 percent of mainline Protestants; data not shown). There are also slight differences among the three major Protestant groups (and a larger difference between Protestants and Catholics) on our religious activity index. The overall sample mean for this index is 17.4. All three Protestant groups have aboveaverage mean scores (21.7 for African-American Protestants, 20.2 for evangelical Protestants, and 19.3 for mainline Protestants), and the mean score for Catholics is below the sample average at 13.9 (data not shown). Thus we might conclude that religiosity tracks Protestantism except for the fact that respondents in the ‘‘other religions’’ category have an average activity score of 17.7, which is also slightly higher than the sample mean (data not shown). We therefore conclude that religious affiliation and religiosity sufficiently measure distinct dimensions of religious life. Despite possible challenges of multicollinearity, we use these measures separately in the multivariate analyses that follow. In fact, when we added various interaction terms to our models, they were not significant. Moreover, all varianceinflation factors for all three analyses were less than 2.3.

In our multivariate analyses, we control for several attitudinal and demographic variables. First, we include a survey item that asks respondents to indicate the top two issues about which they were most concerned. The response set is a prepared list of 12 issues including moral values, terrorism, the economy, education, and Social Security (among others). We created a dummy variable that equals 1 if the respondent listed moral values among his or her top two issue concerns and 0 if the respondent prioritized other issues. We would expect that individuals who prioritize moral values might be more opposed to same-sex unions than those who prioritize other issues, particularly since the phrase ‘‘moral values’’ was used so pervasively in the 2004 political mobilization of evangelical Protestants and traditional Catholics (Green et al., 2004; IVoteValues.com, 2004; Pew Forum on Religion and Public Life, 2004a; Pew Research Center for the People and the Press, 2004). On the whole, 23.5 percent of the sample included moral values among their top two issue concerns, a finding that closely parallels 2004 Election Day exit poll results (Edison Media Research and Mitofsky International, 2004).

Second, we include an item that asks respondents to indicate ‘‘how worried [they] are that society is becoming too secular.’’ Respondents who fear encroaching secularism are likely to believe that traditional family arrangements are being phased out of the public square by secular media, academia, and government (for very different takes on this argument, cf. Neuhaus, 1984; Savage, 2003). We expect that individuals who are worried about societal secularization will oppose same-sex unions because they are likely to see such unions as an assault on traditional values. Table 1 reveals that precisely two-thirds of the sample respondents are at least somewhat worried about secularization.

We also include a general measure of political ideology in our analyses. The survey instrument did not use the standard seven-point ideology scale typically employed by political scientists. Instead, respondents were asked whether they were conservative, moderate, or liberal, with no probes or other means of differentiation. Political conservatives, of course, ought to be most opposed to same-sex unions. As Table 1 shows, 44.8 percent of the sample self-identify as political conservatives.

Our demographic controls include gender, education, age, and marital status. Race is captured in our model via the African-American Protestantism dummy. (We also ran models with controls for income, residence in an urban area, and employment status. These variables were not significant in any of our models, however, so we dropped them.) Previous studies document the fact that gender, education, and age are related to tolerant attitudes toward homosexuality, and marital status might also bear a significant relationship to one’s views on same-sex unions. Table 1 presents the gender, education, and marital-status breakdown of the sample. The average age of the survey respondents was 47.5 years.

Below, we present a multivariate examination of the relationship between religion and attitudes about the three distinct aspects of the dispute over same-sex unions. The question at the center of recent public debates is whether the American public believes that fully sanctioned gay marriages ought to be allowed. Thus we begin by modeling approval of gay marriage. A related question (which frames our second model) is whether civil unions ought to be permitted under law. Support for civil unions might be viewed as either a compromise position or a civil rights issue for people who do not support full gay marriage but nonetheless want same-sex couples to be able to have their relationships legally recognized and respected. Finally, we model support for a federal constitutional amendment to limit marriage to heterosexual couples.

In all three of our analyses, we employ logistic regression, so the dependent variables are dummy dichotomies. For the gay marriage analysis, 1 5 oppose or strongly oppose gay marriage and 0 5 any other attitude. For the civil unions analysis, 15 oppose or strongly oppose civil unions and 0 5 any other attitude. For the federal marriage amendment analysis, 15 support for a constitutional amendment and 05 opposition to such an amendment, reflecting the response set for the question as it was asked. It is worth noting that the items on gay marriage and civil unions are ordinallevel variables, so ordered logit would have been possible in these two cases. However, we are interested in the general direction of respondents’ attitudes about same-sex unions rather than the intensity of these attitudes. As such, we decided to use logistic regression.

Our independent variables include dummies for affiliation with six major religious traditions (African-American Protestantism, evangelical Protestantism, Catholicism, Judaism, other religion, and unaffiliated; the excluded reference category is mainline Protestantism); the religious activity index score; the number of close congregational friendships;3 a dummy where 1 5 listing ‘‘moral values’’ among one’s top two issue concerns; views on whether ‘‘society is becoming too secular’’; and a dummy for political conservatism.4 We also control for gender (a dummy where 1 5 female), education (measured using a six-point scale), marital status (a dummy where 1 5 married), and age (in years).

To compare the relative effects of different independent variables in logistic regression, it is necessary to compute a standardized logistic regression coefficient (which is analogous to beta in ordinary least squares regression analysis), bn (Menard, 2002).5 These standardized coefficients for each logistic regression analysis are reported in Tables 2–4. The interpretation of bn is straightforward: ‘‘a 1 standard deviation increase in X produces a bn 3 The congregational friendships concept is measured on a scale of 0 to 4, where 0 5 none of the respondent’s closest friends are from their congregation (or he or she is not part of a congregation at all); 15 ‘‘some’’ of the respondent’s closest friends are from his or her congregation; 25 ‘‘about half’’; 35 ‘‘most’’; 45 ‘‘all.’’

4 Recall that political ideology was not measured using the standard seven-point scale; instead, respondents labeled themselves as liberal, moderate, or conservative. 5n

The formula for the computation of b is:

bYX ¼ ðbYX ÞðsX ÞðrÞ=slogitðYhatÞ:

In this formula, bYX is the logit coefficient for a given independent variable; sX is the standard deviation for that independent variable; r is equal to the square root of (sY-hat)2 divided by (sY)2 (where sY-hat is the standard deviation of the predicted values of the dependent variable and sY is the standard deviation of the actual values of the dependent variable); and slogit(Y-hat) is the standard deviation of the predicted value of logit (Y), where logit(Y hat) 5 ln[(Y hat)/(1 Y hat)].

standard deviation change in logit(Y)’’ (Menard, 2002:53). Standardized coefficients allow us to assess substantive significance, whereas p values allow us to assess statistical significance.

Tables 2, 3, and 4 report the results of our three logistic regression analyses. The gay marriage model produces a Nagelkerke R2 of 0.42 and a proportional reduction in error (PRE) of 0.28. The civil unions model produces a slightly lower Nagelkerke R2 of 0.36, but a more impressive PRE of 0.38. Finally, the federal marriage amendment model is less powerful; the Nagelkerke R2 is 0.11 and the PRE is 0.15. The relative weakness of the amendment model is likely due at least in part to the fact that only opponents of gay marriage were asked the question about amending the Constitution to prohibit it. Thus there is systematically less religious and attitudinal variance among the respondents in the subsample for this item. Nonetheless, the amendment model enables us to disentangle the strongest opponents of same-sex unions (those who support the Federal Marriage Amendment) from those who are less interested in top-level policy change to prohibit gay marriage.

Nearly every independent variable in the gay marriage model is statistically significant. Only evangelical Protestantism and marital status are insignificant predictors of opposition to gay marriage. It appears counterhypothetical that being an evangelical Protestant is not significantly related to opposition to gay marriage but, instead, being a member of any religious tradition other than evangelical Protestantism (or mainline Protestantism, our reference category) increases one’s likelihood of supporting gay marriage. Comparing the values of bn (the standardized logistic regression coefficients), we find that being Jewish makes one substantially less likely to oppose gay marriage (although a caveat must be offered: the number of Jews in the sample was quite small). Being a member of a religion other than mainstream Christianity or Judaism, or being secular, also render individuals more likely to support gay marriage. Professing concern about moral values, on the other hand, makes individuals substantially more likely to oppose gay marriage. Notice that many of the religious tradition measures and the religious attitudinal measures perform better in our model than the standard ideological and demographic explanations of attitudes toward gay marriage. Notice as well that despite possible multicollinearity concerns, each of the attitudinal variables remains statistically significant on its own. Somewhat surprisingly, very low values of bn attach to the religious practices measures despite the fact that these measures both attain statistical significance in our model.

The civil unions model tells a somewhat similar story, as demonstrated in Table 3. In this model, religious tradition plays a smaller explanatory role in terms of statistical significance than it did in the gay marriage model; only being Catholic contributes significantly to the model, and it does so in what might be seen as a counterintuitive direction (since the Roman Catholic Church officially opposes gay marriage). Despite not attaining statistical significance, the values of bn indicate that being Jewish, a member of a less traditional religious group, or secular again has substantive significance, directing individuals toward support for civil unions (although small-N caveats again apply). Meanwhile, the religious practices measures are again statistically significant but attain low standardized coefficients. As is the case in the gay marriage model, concern about moral values and conservative ideology are significant predictors of opposition to civil unions. The bn values associated with each of these variables show that both contribute substantially to opposition to civil unions. And once again, our religion variables perform better than standard demographic predictors of opposition to civil unions.

The results of the gay marriage and civil unions models lead us to two conclusions. First, religion (especially as measured by religious affiliation and attitudes about morality and secularism) has a powerful effect on attitudes toward same-sex unions. Being a member of a non-Protestant religious tradition appears to lead individuals away from opposition to both gay marriage and civil unions, whereas espousing traditional attitudes on morality and secularism makes individuals more likely to oppose same-sex unions. Second, our strong findings regarding the impact of prioritizing moral values on attitudes toward gay marriage and civil unions remind us of the much-discussed exit poll result from November 2004 (Edison Media Research and Mitofsky International, 2004). Prioritization of moral values above other issues is clearly a significant predictor of opposition to gay marriage. Thus it may well be correct to assume that opposing gay rights is a key component of the rhetorically nebulous notion of ‘‘morality’’ as a political issue.

Our final question is whether opponents of gay marriage also support a federal constitutional amendment that would define marriage as the union between one man and one woman. Theoretically, we could find the strongest relationship between religion and a Federal Marriage Amendment because such an amendment could be seen as the strongest way to prohibit same-sex marriage nationally. However, some conservatives might not support amending the Constitution under any circumstances (see Reeves and Stewart, 2002), so it is somewhat more difficult to generate hypotheses about the factors that should be expected to structure attitudes about a Federal Marriage Amendment.

Recall that the question about amending the Constitution was asked only of respondents who expressed some level of opposition to gay marriage. The item asked whether ‘‘it is enough to prohibit gay marriage by law’’ or if it would be better to ‘‘amend the Constitution.’’ Thus our analysis attempts to identify the factors that drive people to the strongest possible level of opposition to gay marriage.

Here, we see that religious affiliation on its own cannot explain all the nuances of opinion about gay marriage (Table 4). None of the religious tradition dummy variables are statistically significant (although several, especially Judaism, attain substantive significance as measured by bn). To see whether multicollinearity was at work in this model, we included interaction terms between each major religious tradition and religious activity and congregational friendships (data not shown), but these did not produce significant results. The only significant finding regarding any religion variable is that high levels of religious activity contribute to support for a constitutional amendment, although the substantive significance, as measured by bn, for this variable is again rather small. Among the statistically significant predictors in the model, political conservatism is by far the strongest predictor of support for a Federal Marriage Amendment. As such, we conclude that while religion plays strong roles in structuring attitudes toward gay marriage and civil unions, it does not help us differentiate between those who wish to amend the Constitution to prevent same-sex unions and those who prefer statutory prohibitions. Being highly involved in religious life thus promotes opposition to same-sex unions, but it does not go particularly far in helping us understand people’s attitudes about how government should prohibit such unions. Additional analyses of public opinion about amending the Constitution—using data that include information about political party affiliation from a much wider range of respondents than those who were asked the question in this survey—will be important lines for future research.

Religion, as measured both by religious affiliation and religiosity, has a powerful effect on public opinion about same-sex marriage and related issues in the United States. In particular, non-Protestants are much more likely to support same-sex unions, and individuals with conservative attitudes on morality and secularism and (to a lesser extent) those who participate actively in religious life are more likely to oppose such unions. The fact that our religion variables perform better than demographic measures in our models clearly shows how important religion is in the shaping of attitudes about same-sex unions.

Further, despite small values of bn, our findings about religiosity—specifically the statistically significant relationship between congregational friendships and opinion about same-sex marriage—point to the ways informal friendship networks in religious contexts might enhance opposition to gay rights. People with many close friends in their religious congregation are most enmeshed in their congregation. The more friends a person has in his or her congregation, the more he or she will be invested in the congregation’s future, and the more his or her sociopolitical viewpoints will be structured by the consensus of the congregation, or at least the consensus of the friendship network within the congregation. Congregations, regardless of their specific religious affiliation, can be highly cohesive political communities in which opinion tends to converge on certain issues, particularly those that are morally charged (Gilbert, 1993; Wald, Owen, and Hill, 1988, 1990). This is the case even though many clergy are hesitant to preach or speak publicly about homosexuality for fear of alienating congregation members (Damore, Jelen, and Bowers, forthcoming; Olson and Cadge, 2002).

Our results also offer a suggestion that Americans who profess concern about moral values (as did 22 percent of the American electorate on Election Day (Edison Media Research and Mitofsky International, 2004)) are motivated to do so at least in part by their opposition to gay rights. Our analyses show that identifying moral values among one’s top two issue concerns is a strong and significant predictor of opposition to both gay marriage and civil unions. Moreover, the Republican mobilization effort directed toward evangelicals and traditional Catholics emphasized gay marriage as an issue of crucial importance (Green et al., 2004; IVoteValues.com, 2004)—and attracted many of the faithful across religious traditions to vote for President Bush (Green et al., 2004; Pew Forum on Religion and Public Life, 2004a; Pew Research Center for the People and the Press, 2004).

More broadly, our results suggest that without a change in opinion among religious individuals and organizations in the United States, the tide is not likely to turn in favor of same-sex marriages or civil unions without some reframing of the issue. Battles about homosexuality are being waged in religious organizations across the country, although this has been the case to a greater extent in mainline Protestant and Jewish circles than in other religious traditions, most notably evangelicals (Cadge, 2002; Zuckerman, 1999). It is useful to remember that American religion is not uniformly opposed to same-sex unions (Cadge, 2002), but opponents do vastly outnumber supporters. Congregations are important crucibles of public opinion, particularly on charged moral issues. It is clear from our analysis that religious context has a powerful effect on citizen opinion about same-sex unions.

1 The response rate for the survey was 15 percent. The response rate is low because the primary intention of the survey was to oversample evangelical Protestants, African Americans, and Hispanics. The survey methodology used by Greenberg Quinlan Rosner was similar to that used by Gallup, Princeton Survey Research Associates, the Pew Research Center, and other major polling firms, involving multiple callbacks, refusal conversions, and weighting, and followed AAPOR guidelines. An examination of the demographic characteristics of the sample led us to conclude with confidence that the sample is representative (data not reported).