Voting to Ban Same-Sex Marriage: Interests, Values, and Communities

From 2000 through 2008, initiatives proposing to ban same-sex marriage were on the ballot in 28 states. Although same-sex marriage opponents scored lopsided victories in most cases, voting outcomes varied substantially at the county level. This article examines sources of that variation and argues that opposition to same-sex marriage should be strong in communities characterized by the predominance of traditional gender roles and family structure. Perhaps more interestingly, the analysis also shows that the effects of traditional family structure and gender roles are especially strong in counties characterized by weak community cohesion, as indicated by residential instability, low rates of home ownership, and high crime rates.

one woman" and specifies that states do not have to recognize same-sex marriages granted in another state (Cahill 2004). This federal legislation passed with nearly unanimous support from Republican legislators and majority support from Democrats. Democratic President Bill Clinton signed the bill into law. DOMA was followed by a flurry of state-level activity, as legislators sought to ban same-sex marriage or to reinforce language in preexisting laws that made same-sex marriage illegal (see Soule [2004] for an analysis of the timing of state-level legislative bans on same-sex marriage from 1973 to 2000).
From 1998 to 2008, ballots in 30 states had initiatives to ban same-sex marriage. As Table  1 shows, opponents of same-sex marriage scored lopsided victories in most cases. Indeed, a same-sex marriage ban was defeated in only one state, Arizona (and by a very narrow margin). Only two years later, Arizona voters revisited the issue and approved a constitutional amendment banning same-sex marriage. Although these voting outcomes reflect broad opposition to same-sex marriage throughout the U.S. population, state-level figures obscure substantial variation in how local communities have responded. Examining the voting data at the county level, rather than the state level, we find that support for a same-sex marriage ban ranges from a low of 23 percent in the county equivalent of Charlottesville, Virginia to a high of 95.4 percent in Martin County, Texas.

Voting to Ban Same-Sex Marriage: Interests, Values, and Communities
Voting data derived from these ballot initiatives provide a unique opportunity to examine sources of support for, and opposition to, samesex marriage. The issue is important in its own right because it involves a conflict over basic civil liberties, and the ultimate outcome of the conflict has implications for millions of lesbians and gay men in the United States. Both instrumental and symbolic goals are at stake. Supporters of same-sex marriage seek equal access to rights and privileges that heterosexual married couples currently enjoy. Perhaps just as important from the perspective of gay rights supporters, legalizing same-sex marriage would mean the government could no longer enforce restrictions on marriage based on an understanding of homosexual unions as deviant and illegitimate.
More generally, the voting data provide an opportunity to examine how local communities' structural features shape public opinion on an important issue of contention. The fight over same-sex marriage represents a competition over socially constructed meanings, as groups and individuals on both sides of the issue struggle to define the meanings of marriage, sexuality, morality, and civil rights, and to assert their own definitions of the situation in the face of organized opposition. In our analysis, we focus on communities (with counties as 892--AMERICAN SOCIOLOGICAL REVIEW proxies) rather than individuals because opinions on same-sex marriage are formed through social interaction. Varying understandings of same-sex marriage develop through life-long socialization processes and are constructed and reconstructed in everyday conversations in churches, college campuses, coffee shops, workplaces, and bowling alleys. The nature of these conversations, and the ways in which various arguments about same-sex marriage resonate with individuals, should depend on varying patterns of social relations across local contexts. We identify structural features of local settings that affect the extent to which same-sex marriage is perceived to be threatening to (1) community residents' interests and values and (2) the community as a whole. As Eskridge and Spedale (2006) note, some opponents of same-sex marriage express their opposition simply on definitional grounds. Marriage is understood as a union of one man and one woman, and no other type of arrangement, from this perspective, should be permissible under law. Increasingly, however, opponents of same-sex marriage have articulated a consequentialist argument, claiming that same-sex marriage is not only wrong on moral grounds but would also harm society (Eskridge and Spedale 2006). We argue that traditional gender roles and family structure promote opposition to same-sex marriage and that these effects should be particularly strong in locations where many residents are not deeply rooted in the community. Traditionalism, we propose, leads many community residents to view homosexuality as a threat to their interests and values. In communities where many individuals feel threatened by homosexuality, weak community cohesion can contribute to a general sense that same-sex marriage would further undermine the community's social fabric.
When we use the phrase "weak community cohesion," we are thinking of circumstances in which many community residents do not feel a strong attachment to their neighborhoods, towns, or cities and are likely to be unfamiliar with many individuals with whom they have contact throughout the day. These conditions, as Durkheim (1933) and Simmel (1950) pointed out long ago, while weakening pressure toward conformity, can also undermine trust, solidarity, and a commitment to the collective good.

THREAT TO INTEREST AND VALUES
Although attitudes about gay rights are becoming increasingly liberal, a sizeable proportion of Americans oppose same-sex marriage simply because they oppose homosexuality. Certainly, many Americans continue to reject same-sex marriage on definitional grounds and do not feel compelled to offer further justification. We expect that these attitudes are clustered within communities characterized by traditional family structures and gender roles. As Jackman (1994:59) argues, groups tend to gravitate "to ideologies that are rationally consistent with their interests," and we believe the same is true in regard to how they form opinions regarding same-sex marriage. This does not mean that all opponents of same-sex marriage adhere to a single coherent and systematic set of beliefs, or that they deliberately promote such a belief system to gain a strategic advantage for their group. Instead, we think that arguments that speak of the importance of preserving traditional marriage are more likely to "ring true" in communities where traditional gender and family arrangements predominate.
Psychological studies show that, at the individual level, prejudiced attitudes toward homosexuals are correlated with adherence to traditional sex roles (Basow and Johnson 2000;Cotton-Huston and Waite 2000;Polimeni, Hardie, and Buzwell 2000). 1 Individuals who adhere to traditional sex roles typically develop a stake in maintaining the status quo. They may view lesbians and gay men as economic threats, as some heterosexual individuals believe that gay rights will lead to special treatment or reverse discrimination (Bernstein, Kostelac, and Gaarder 2003;Stein 2001). Heterosexual men, who derive a broad array of benefits by embodying masculine traits, may feel that homosexuality represents a threat to male privilege (Bernstein 2004). Many women and men who adhere to traditional gender roles may also believe that same-sex marriage undermines the institution of marriage and devalues the status associated with heterosexual marriage VOTING TO BAN SAME-SEX MARRIAGE--893 (Bernstein 2004). We expect opposition to samesex marriage to be strong, therefore, where a large proportion of the population has a personal stake in maintaining traditional gender roles and family structures.
Yet the geographic concentration of opposition to same-sex marriage represents more than a simple clustering of individual traits and predispositions. Britton (1990), for example, argues that negative attitudes toward homosexuals are rooted in sex-segregated social arrangements. Drawing on the ideas of Lipman-Blumen (1976), Britton (1990:425) proposes the following: A stratification system in which males and all-male institutions have almost exclusive access to major resources reinforces same-sex interest. Men homosocial in outlook prefer other men's company and also work to maintain all-male institutions. The relationship to homophobia lies in maintaining the boundary between social and sexual interaction in a homosocially stratified society.
According to this logic, traditional gender roles and family structures should foster negative attitudes toward both homosexuality and same-sex marriage due to the prevalence of social norms and practices that reinforce boundaries and maintain sex-based power differentials.
Other research calls attention to how network ties to lesbians or gay men tend to reduce sexual prejudice among heterosexuals (Bernstein et al. 2003;Herek and Capitanio 1996). In the absence of such ties, negative stereotypes and myths pertaining to homosexuality may go unchallenged (Bernstein 2004;Herek 1991). In communities where traditional gender roles and family structures predominate, individuals are less likely to have contact with people who openly express a gay or lesbian identity. In this context, individuals should be more likely to hold negative stereotypes and to view homosexuality as immoral and threatening to the privileges they enjoy.

THREAT TO COMMUNITY
Consequentialist arguments against same-sex marriage characterize traditional marriage as an "altruistic space" wherein adults subordinate their own self-interest to their children's welfare. This altruistic space serves an important public function, "as it trains good citizens, rears superior children, and creates greater happiness in the long run" (Eskridge and Spedale 2006:28-29). The argument attributes numerous social ills to the decline of the traditional family in modern society and interprets same-sex marriage as a serious threat that will further undermine the family unit's altruistic functions. For example, Blankenhorn (2007:96-97) argues that acceptance of samesex marriage necessarily undermines marriage as a social institution, which is, he asserts, "not a 'bundle of rights,' but a pattern of rules and structures intended to meet societal needs." Stein's (2001) study of opposition to gay rights in a small Oregon town shows how conservatives use consequentialist arguments against same-sex marriage. Community activists opposed gay rights on religious grounds, but they recognized that to expand their base of support they would need to appeal to individuals leery of imposing one set of religious values on all citizens. Stein calls attention to how conservatives tapped into a general sense that gay rights pose a threat to an already fragile community cohesion. The mayor of the small Oregon town, for example, lamented: "It used to be that people socialized in groups. Today, people are more isolated. There's less interconnectedness" (Stein 2001:64).
In a study of attitudes about homosexuality, Persell, Green, and Gurevich (2001:213) argue that a breakdown in civil society undermines trust and promotes intolerance. While Persell and colleagues studied individuals' attitudes, their findings are consistent with Stein's observations about opposition to gay rights in Oregon. As Stein (2001:215) writes, a general sense that social cohesion was weakening led many in the community to fear the "strangers" in their midst: "Talk of community surfaces most when communities are under threatand frequently leads people on quests to repair them." Opponents of same-sex marriage view the traditional family as the primary institution for socializing children and for maintaining social order and community cohesion. Samesex marriage is seen as a threat to traditional marriage and, by extension, to moral and cohesive communities. Florida Congressman Charles Canaday exemplified this view when arguing on behalf of the Defense of Marriage Act in 1996: What is at stake in this controversy? Nothing less than our collective moral understanding-as expressed in the law-of the essential nature of the family-the fundamental building block of society. This is far from a trivial political issue. Families are not merely constructs of outdated convention, and traditional marriage laws were not based on animosity toward homosexuals. Rather, I believe that the traditional family structure-centered on a lawful union between one man and one womancomports with nature and with our Judeo-Christian moral tradition. (The Congressional Record 1996) In this perspective, by weakening the fundamental building block of society, same-sex marriage would contribute to, or exacerbate, the breakdown of social order.
Organized opponents of same-sex marriage often express these types of concerns when justifying their intent to restrict gay and lesbian individuals' civil liberties. On its Web site, the Family Research Council (2007), for example, makes its case against same-sex marriage by referring to research carried out by "a group of thirteen leading social scientists." The reference is to a report published by the Institute for American Values (2002:18), which concludes: Marriage is more than a private emotional relationship. It is also a social good. Not every person can or should marry. And not every child raised outside of marriage is damaged as a result. But communities where good-enough marriages are common have better outcomes for children, women, and men than do communities suffering from high rates of divorce, unmarried childbearing, and high-conflict or violent marriages.
Not surprisingly, same-sex marriage opponents neglect to call attention to the fact that scholars are divided on whether changes in family structure lead to a decline in positive social outcomes. 2 Yet even if one takes such a report at face value, it could just as easily be used to advocate for, rather than against, same-sex marriage. If marriage is a public good, why limit it to heterosexual couples? The Family Research Council (2007) responds to this question by returning to the definitional argument against same-sex marriage: The benefits of marriage do not simply flow from the presence of two people and government recognition of their relationship. Instead, they flow from the inherent complementarity of the sexes and the power of lifelong commitment.
Only those who accept the initial premisethat homosexual unions defy the laws of nature and the laws of God-would f ind such a response satisfying. For such individuals, however, the consequentialist argument provides justification for restricting gays' and lesbians' rights because they interpret the preservation of traditional marriage as a public good.
We do not expect to find that weak community cohesion, by itself, generates opposition to same-sex marriage. Indeed, weak cohesion that results when many residents are not firmly rooted in a particular neighborhood, town, or city can have a liberalizing effect. Anonymity relieves pressure toward conformity (Durkheim 1933;Fischer 1995;McVeigh 1995;Simmel 1950). In many locales, weak community cohesion is not broadly viewed as problematic and would not provoke opposition to same-sex marriage; even if community members view weak cohesion as a problem, they would not regard banning same-sex marriage as the solution. However, traditional sex and gender roles within a community can promote opposition to same-sex marriage in a way that leads many to see a breakdown in social organization as both a cause and a consequence of nontraditional values, including values that promote tolerance of same-sex marriage. We therefore expect to find strong opposition to same-sex marriage in communities where traditional family structures and gender roles are most prevalent, and the effects of these community attributes should be particularly strong in settings characterized by weak community cohesion.

DATA AND METHODS
To test our argument, we examine county-level voting results in the 28 states that had same-sex marriage initiatives on the ballot from 2000 to 2008. Because most of our independent variables are measured using data from the 2000 U.S. Census, we do not include earlier votes in Alaska and Hawaii. For Arizona and California, we use the results of earlier elections, rather than the most recent elections in 2008, because the earlier elections were conducted closer to the VOTING TO BAN SAME-SEX MARRIAGE--895 period used to measure our independent variables. 3 Our dependent variable is the percent of votes cast in each county in support of the samesex marriage ban. In all but one of the states in the analysis, the ballot initiatives proposed to amend the state constitutions so that marriage would be explicitly limited to a union between one man and one woman. California's vote in 2000, unlike the more recent Proposition 8 in that state, banned same-sex marriage but did not amend the state constitution. In Nevada, the issue appeared on the ballot in both 2000 and 2002 (in Nevada, an initiative to amend the state constitution must be approved in two consecutive elections before it takes effect). In our analyses, the measure of opposition to same-sex marriage in Nevada's counties represents the average percentage of voters approving the ban across the two election periods.
We are interested in identifying ways in which attributes of local contexts influence the vote on same-sex marriage. We use counties as units of analysis because a state-level analysis would ignore substantial intrastate variation in our relevant variables. Conditions experienced by individuals in one part of a state can be quite different from those experienced by residents elsewhere in the state, and these differences could be associated with attitudes toward samesex marriage. However, some of the variation across counties is certainly attributable to statelevel differences. States phrase the ballot initiative questions differently and also differ in their openness to alternatives to same-sex marriage, such as state-sanctioned civil unions. In addition, the elections occurred across an eightyear time span, and the year in which an election was held could be related to the results. Failure to control for these and other state-level effects could produce biased estimates. To address this problem, we use a fixed-effects model estimated with the AREG command in Stata. The procedure, in effect, inserts a dummy variable for each of the 28 states included in our analysis. Our regression results thus represent the estimated effects of our independent variables on opposition to same-sex marriage after controlling for state-level differences in voting. We also use robust standard errors for significance testing to correct for unmeasured but stable state-level factors that may cause correlated errors among observations within each state.

TRADITIONAL FAMILY STRUCTURE AND GENDER ROLES
Opposition to same-sex marriage should be strong in communities where traditional family structure and gender roles are predominant. We include two measures related to labor force participation: (1) the percentage of women age 16 years and older who reported that they did not work in 1999 and (2) a measure of occupational sex segregation. The 2000 Census breaks county residents' occupations into 93 groupings. We calculated an index of dissimilarity with the value representing the proportion of either men or women who would have to change occupational categories to produce an even distribution of men and women across categories, where sex is not correlated with occupation (McVeigh and Sobolewski 2007).
We also include several variables related to family structure, including the percentage of households in a county composed of a married couple with children under age 18. Of course, we do not expect that all individuals in this traditional family arrangement will oppose samesex marriage. At the county level, however, we expect to find that opposition to same-sex marriage will be strongest where this traditional family form is normative. We also include variables capturing variation in the presence of nontraditional families, including the percentage of households in a county composed of a householder and a same-sex partner. This is a flawed measure if the goal is to accurately assess the prevalence of gay and lesbian households. Many individuals, when filling out the short form of the Census, are reluctant to reveal the nature of their relationship to a same-sex partner, especially in communities where prejudice against gays and lesbians is strongest. For our purposes, however, the measure does help identify county-level differences in the general openness to same-sex relationships. Values for this vari-

896--AMERICAN SOCIOLOGICAL REVIEW
able represent some combination of the presence of same-sex households and a willingness to reveal such relationships when filling out Census forms. Where the percentage of samesex households is low, we expect to find stronger opposition to same-sex marriage. Using the same Census table, we also calculate a measure of the percentage of households with either a male householder and an unmarried female partner, or a female householder with an unmarried male partner. This measure, similar to the measure of same-sex households, does not necessarily provide an accurate tally of the proportion of households made up of cohabiting couples, but it does reflect a general openness to nontraditional heterosexual households.

WEAK COMMUNITY COHESION
We include several variables related to community cohesion. Social disorganization theory proposes that residential instability can disrupt social organization in communities and loosen constraints on social behavior (Sampson and Raudenbush 1999;Shaw and McKay 1942). Residential instability, in this sense, can reduce pressure toward conformity and have a liberalizing effect on community residents. However, we expect that in communities where traditionalism is most prevalent, these same conditions increase the chance that community residents view same-sex marriage as threatening to both the community and their own interests and values. Using the 2000 Census, we measure residential instability as the percentage of individuals age 5 years and older who are living in a residence different from the one in which they lived five years earlier. Low rates of home ownership can also be a source of weak community cohesion (Krivo and Peterson 2000), so we include the percentage of homes in a county that are not owner-occupied.
We also include a measure of the crime rate. Crime, of course, is a frequent topic of study for social scientists. Rarely, however, do social scientists treat crime as an independent variable. Criminological research and theory suggest that crime can result from social disorganization (Bursik and Grasmick 1993;Sampson and Laub 1993;Shaw and McKay 1942). In the absence of strong social bonds connecting members of a community, individuals face fewer social constraints on criminal and delinquent behavior.
Crime, in other words, results from a breakdown in the social relations and common values that sustain healthy communities. Often left unsaid is that a high crime rate is itself an indicator of a breakdown in community cohesion and solidarity (see Liska, Logan, and Bellair 1999;Quillian and Pager 2001;Rountree and Land 1996).
Opponents of same-sex marriage sometimes argue that legalizing same-sex marriage will contribute to higher crime rates. Most of these arguments, however, do not suggest that gay individuals will be the ones committing these crimes. Instead, opponents argue that legalization will undermine the institution of marriage and lead to poorer socialization of children. We cannot say how widely accepted such beliefs are among voters. In our analysis, we see crime operating more as an additional indicator of weak community cohesion that should be similar to residential instability and low rates of home ownership in terms of how it affects voting outcomes. A high crime rate indicates that many individuals in a community feel unconstrained by ties to the community or its residents.
We constructed a measure of the crime rate for the year 2000 using data provided by the Federal Bureau of Investigation's Uniform Crime Reports (UCR). We accessed the county-level data through the Web site maintained by the Inter-university Consortium for Political and Social Research (ICPSR). We calculated our measure as the number of index crimes (i.e., murder, rape, robbery, aggravated assault, burglary, larceny, motor vehicle theft, and arson) for every 1,000 residents reported for each county. Unfortunately, concerns about data quality compelled us to exclude 629 counties when including the crime variable in our analyses. Therefore, we estimate most of our models without the crime variable but then estimate the effects of the crime rate in separate models with a more limited sample. See McVeigh (2006) for a detailed discussion of the county-level UCR data. 4

CONTROL VARIABLES
Some researchers argue that opposition to samesex marriage, and to homosexuality more generally, is most likely found in communities where residents are struggling economically. Persell and colleagues (2001), for example, argue that economic hardship can undermine civil society and trust and can decrease levels of social tolerance. Other research suggests that gay rights might be perceived as threatening in working-class and less prosperous communities where many residents may be receptive to arguments about how policies designed to protect or promote gay rights would lead to reverse discrimination (Bernstein et al. 2003;Stein 2001). Theories of status politics and status substitution (Gusfield 1963;Lipset and Raab 1970) suggest that support for conservative values represents a form of low-status backlash. We control for features of the economic structure that might be related to a vote in opposition to same-sex marriage. The 2000 Census groups occupations into six broad categories. To form a measure of working-class or blue-collar employment, we calculate the percentage of workers employed in either of the following two general categories: (1) construction, extraction, and maintenance and (2) production, transportation, and material moving occupations. We expect opposition to same-sex marriage to be stronger in counties where high percentages of workers are employed in these blue-collar occupations. We construct another measure of the percentage of individuals employed in professional occupations, which we expect to be negatively related to the dependent variable. We also include a measure of the percentage of households receiving self-employment income, because the self-employed may be less likely to be concerned about economic competition that could result from an expansion of gay rights.
We also control for counties' median income in 2000. As an additional indicator of economic conditions, we use Census data to obtain a measure of the percentage of households receiving public assistance income in 1999. Research consistently shows that education tends to produce more liberal attitudes regarding homo-sexuality and gay rights (Herek and Glunt 1993;Loftus 2001;Treas 2002). In recent decades, increasing tolerance of homosexuality in the population at large can be attributed, in part, to increases in the number of individuals pursuing higher education (Loftus 2001). We use 2000 Census data to obtain a measure of the mean years of education for county residents age 25 and older. Above and beyond the effects of education, many college campuses provide contexts conducive to gay rights activism and more tolerant views about homosexuality (Wald, Button, and Rienzo 1996). To capture this "college town" effect, we include a measure of the percentage of county residents age 3 years and older who are currently enrolled in college. 5 Responses to same-sex marriage are also likely related to urbanization and population density. Using Census data for the year 2000, we calculate the natural log of the population per square mile. We also include a measure of the percentage of county residents living in an urban location. We expect to find stronger opposition to same-sex marriage in counties where conservative belief systems are broadly accepted. Therefore, we include a measure of the percentage of county voters who voted for Republican presidential candidate George W. Bush in the 2000 election. Party loyalty and partisanship, as Campbell and colleagues (1960) noted long ago, can act as a lens through which all issues are viewed, interpreted, and understood. Bush, and more generally the Republican Party, appealed to voters by emphasizing their support of traditional family values, and these appeals solidified support for the party among socially conservative voters (Coltrane 2001;Oberschall 1993;Regnerus, Sikkink, and Smith 1999;Rudolph and Grant 2002).
Political advocacy of traditional family values has, to a great extent, been targeted toward Evangelical Protestants, and several leaders of Evangelical organizations and churches are outspoken opponents of same-sex marriage. Evangelical Protestants are not homogeneous in

898--AMERICAN SOCIOLOGICAL REVIEW
county only if the population served by the reporting agencies represents at least 90 percent of the total county population.
terms of their beliefs about social issues, and as a group they do not differ significantly from other Americans on many political issues. However, they do tend to be particularly conservative on issues related to homosexuality and gay rights (Smith 2000).
We use data from the American Religion Data Archive (ARDA) to calculate a measure of Evangelical Protestants as a percentage of all church adherents in each county in the year 2000. ARDA's classification of denominations as Evangelical is guided primarily by Steensland and colleagues' (2000) coding strategy and uses information provided in the Encyclopedia of American Religion and the Handbook of Denominations in the United States to classify the denominations not covered (ARDA 2005). We use the same data source to calculate a measure of the percentage of church adherents who are Catholic. The Catholic Church teaches that any sexual act committed outside of heterosexual marriage is sinful and characterizes homosexuality as a disorder (Loseke and Cavendish 2001). Yet U.S. Catholics tend to be more liberal than conservative Protestants on issues related to gay rights (see Bendyna et al. 2001;Finlay and Walther 2003;Haider-Markel and Meier 1996;Kellstedt et al. 1994). In fact, data from the 2006 General Social Survey show that while a minority of Catholics (34 percent) agreed that homosexuals should have a right to marry, only 26.3 percent strongly disagreed with that statement, compared with 47.5 percent of Protestants.
We include a measure of median age in a county because prior research shows that older individuals tend to have more conservative views about homosexuality (Loftus 2001). It is important, however, to keep in mind that a relatively high median age does not necessarily reflect high concentrations of elderly residents; it may instead reflect a low fertility rate. We also control for the percent of a county population that is African American or Latino. Although African Americans and Latinos tend to align themselves with the Democratic Party, they also tend to be more conservative than White Democrats when it comes to views on homosexuality. 6 When considering the 2,231 counties included in our analysis, our data show that opposition to same-sex marriage is weakly correlated with our measure of percent Black (r = .125) and percent Latino (r = .047). We expect these relations to change when we control for state-level differences and other factors related to voting on same-sex marriage.
As Soule (2004) demonstrates, activist organizations can influence public policy on issues such as same-sex marriage (see also Burstein 1991;McVeigh, Welch, and Bjarnason 2003;Wald et al. 1996;Werum and Winders 2001). Soule's analysis uses state-level measures of the presence or absence of interest groups involved in the same-sex marriage debate. Our fixed-effects models are designed to control for these and other state-level differences. At the county level, we include a dichotomous variable coded 1 if a county has at least one lesbian, gay, bisexual, or transgender organization listed under the "Act Locally" link on the National Gay and Lesbian Task Force's Web site. We also use data collected by the National Center for Charitable Statistics (NCCS) to calculate the number of nonprofit organizations coded from R20 to R30 using the National Taxonomy of Exempt Entities coding scheme. These categories consist of activist organizations engaged in civil rights activism for specific groups and organizations that aim to promote increased harmony between groups. NCSS compiled the data set using Internal Revenue Service records for nonprofit organizations filing for tax exempt status each year. 7 Finally, we include a dichotomous variable coded 1 if a county (or a city within the county) enacted legislation prohibiting discrimination in private employment based on sexual orientation or gender identity prior to 2000.

INTERACTION EFFECTS
We argued that opposition to same-sex marriage should be strong in counties characterized VOTING TO BAN SAME-SEX MARRIAGE--899 by traditional gender roles and family structure because many residents in these counties are likely to view same-sex marriage as a threat to their interests and values. We expect the effects of traditionalism to be strongest in counties with weak community cohesion, because under these circumstances community residents are likely to perceive same-sex marriage as threatening not only to their own interests and values but to the community at large. We examine several interaction effects to test this argument. To facilitate interpretation of the interaction effects, we center all variables on their mean values.

DATA ANALYSIS
Our findings can be generalized only to counties in states where initiatives pertaining to same-sex marriage have been on the ballot. To date, 30 states have had such initiatives. Geographic region and political party alliances seem to be important factors in determining which states have had same-sex marriage initiatives. With the exception of North Carolina, all of the southern states have voted to ban same-sex marriage. Of the 20 states that have not had same-sex marriage initiatives on the ballot, 18 gave majority support to Democrat Barack Obama in the 2008 presidential election (West Virginia and Wyoming are the exceptions). 8 The measures for all of our independent variables, as well as the data sources, are described in the Appendix. Univariate statistics for all variables and a matrix of Pearson correlation coefficients are presented in Table A1 in the Appendix. This table reveals some strong correlations among our independent variables. In preliminary analyses, we estimated all of our models with Ordinary Least Squares regression and examined condition index values to detect potential multicollinearity problems. Belsley, Kuh, and Welsch (1980) suggest that condition index values ranging from 30 to 100 indicate moderate to strong multicollinearity, and under such conditions there is a high likelihood that multicollinearity is harming or degrading the regression estimates (see also Podolny, Stuart, and Hannan 1996:679-80;Ross 1987:262). Even with state-level dummy variables included, the highest condition index value in any of our models is 18.2. Table 2 presents results of multivariate analyses. 9 The first two columns show results when state-level effects are not controlled. The first model excludes the crime rate variable and therefore includes 2,231 counties. The second model includes the crime variable. Because of concerns about data quality, models including the crime variable are limited to 1,602 counties. In both models, the independent variables explain over 70 percent of county-level variation in voting. Without controlling for statelevel effects, however, the coefficients could be quite misleading if the goal is to show how county-level attributes are related to voting outcomes. Columns 3 and 4 show results with state-level dummy variables included. In almost all cases, the coefficients of the state-level variables are negative and statistically significant, indicating that after controlling for numerous county attributes, Texas counties have higher levels of opposition to same-sex marriage than do counties in other states.
After controlling for these state-level differences, we find that opposition to same-sex marriage tends to be higher in counties with low percentages of women working in the labor force, high levels of occupational sex segregation, and high percentages of households made up of married couples with children. Also as expected, opposition tends to be lower in counties with higher percentages of nontraditional 900--AMERICAN SOCIOLOGICAL REVIEW 8 We calculated all of our variables at the state level for all 50 states and compared the mean values of states that have had same-sex marriage initiatives with those of states that have not. We found that the mean values of the following variables are significantly higher in states that have had a ballot initiative: Republican voting, crime rate, percent Evangelical, and residential instability. For these same states, the mean values are significantly lower for the following variables: percent Catholic, median age, median income, percent professional occupations, and percent unmarried, opposite-sex households. 9 To obtain the results presented in Table 2, we estimated models using Ordinary Least Squares regression. We manually inserted state-level dummy variables so that state-level effects could be observed and compared. In subsequent tables, we estimated fixed-effects models using the AREG command in Stata. households, as indicated by higher percentages of same-sex partner households and higher percentages of unmarried, opposite-sex-partner households.
After controlling for other variables, residential instability and low rates of home-ownership have a negative effect on the dependent variable. As we argued above, weak community cohesion should not, by itself, generate opposition to same-sex marriage, as weak ties can reduce pressure toward conformity and promote greater appreciation for individuality (Durkheim 1933;Simmel 1950). We argue that these conditions contribute to opposition in locations where traditional family structure and gender roles shape residents' perceptions in a way that makes them likely to view same-sex marriage as a threat to both their own interests and values and the community. However, net of other variables, high crime rates are associated with greater opposition to same-sex marriage.
Among our control variables, opposition to same-sex marriage tends to be stronger in less prosperous counties. Opposition is stronger in counties with high percentages of blue-collar occupations and lower in counties with high percentages of professional occupations and households receiving self-employment income. Opposition also tends to be lower in counties with a high median income, high levels of educational attainment, and high percentages of residents enrolled in college. Our measure of public assistance income, however, falls short of statistical significance.
Although bivariate correlations indicate that opposition to same sex marriage is weaker in densely populated and urbanized counties, when other controls are included in the model the population density measure is not statistically significant, and the measure of urbanization is positive and significant in models using the full set of cases. As expected, Republican voting and our measure of Evangelical adherents have positive and significant effects on the dependent variable, although the measure of Evangelical adherence falls short of statistical significance in the models that include the crime variable with a limited sample of observations. Counties with high percentages of Catholics show less opposition to same-sex marriage. After controlling for state effects and other variables, median age is negatively related to the dependent variable. We also find less opposition to same-sex marriage in counties with high proportions of Latino residents, but only when state-level effects are controlled. Counties with at least one LGBT organization show less opposition to same-sex marriage, but our more general measure of the presence of civil rights organizations (not specifically focused on LGBT concerns) is not a significant predictor of the voting outcome. Not surprisingly, counties that enacted antidiscrimination legislation based on sexual orientation tend to show less opposition to same-sex marriage.
Results presented in Table 2 are consistent with our expectation that opposition to same-sex marriage should be stronger in counties characterized by traditional family structure and gender roles. These findings, while important to understanding opposition to (and support for) same-sex marriage, do not directly test our theoretical argument. We argued that the effects of these measures of traditionalism should be especially strong in counties characterized by weak community cohesion. In such counties, high proportions of residents are likely to view samesex marriage as threatening to both their own interests and values and the community at large. We test these claims by examining a series of interaction effects.
We first examine how residential instability interacts with our measures of traditional family structure and gender roles, while controlling for state-level differences and numerous other relevant variables. As column 1 of Table 3 shows, the coeff icient for the interaction between residential instability and the percentage of women not in the labor force is highly significant. The coefficient for the main effect of the labor force participation variable indicates that when residential instability is at its mean value, the percentage of women not working in the labor force has a significant and positive effect on opposition to same-sex marriage. The positive sign for the interaction term indicates that the effect of our labor force participation variable is stronger in counties with higher levels of residential instability.
We find that occupational sex segregation plays a particularly strong role in shaping voting outcomes. The interaction between sex segregation and residential instability is positive and highly significant. The main effect of the sex segregation variable indicates that when residential instability is at its mean value, sex seg-regation has a strong, positive effect on opposition to same-sex marriage. A one-unit increase in segregation, which represents a transition from a completely integrated labor market to a completely segregated labor market, produces an expected 9.2 percent increase in votes opposed to same-sex marriage. This effect is significantly stronger in counties with high levels of residential instability. Figure 1 depicts this relationship, showing the predicted effect of sex-segregated occupations on the vote against same-sex marriage at varying levels of residential instability.
We observe similar patterns when we interact residential instability with our measures of family structure. Results in column 3 of Table  3 indicate that the coefficient for the interaction of residential instability and percent married with children is positive and significant. In this case, the main effect for the married with children variable is not significant, indicating that percent married with children produces greater opposition to same-sex marriage only in counties with higher-than-average levels of residential instability. As expected, we also find that the negative effect of same-sex households on oppo-VOTING TO BAN SAME-SEX MARRIAGE--903   sition to same-sex marriage is significantly stronger in counties with higher levels of residential instability. Or, looking at this relation another way, having relatively few same-sex households leads to more opposition to samesex marriage, especially in counties characterized by high levels of residential instability. We obtain similar results when we interact residential instability with our measure of oppositesex-cohabiting households. In support of our argument, we find that the effect of traditional gender roles and family structure on voting outcomes is particularly strong in counties characterized by high rates of residential instability. If our theoretical explanation for these findings is sound, we should obtain similar results when we interact our traditional gender role and family structure variables with other indicators of weak community cohesion. Results in Table 4 show how low rates of home ownership combine with traditional gender roles and family structure to promote strong opposition to same-sex marriage. The Table 4 results are strikingly similar to those presented in Table 3, where we considered the role of residential instability. As was true of resi-dential instability, the positive effects of low percentages of women in the labor force and of occupational sex segregation, as well as high percentages of households composed of a married couple with children, are strongest in counties with low percentages of owner-occupied housing. Likewise, the negative effects of samesex households and unmarried, opposite-sex households are strongest in counties with low rates of owner-occupancy.
As Table 5 indicates, the crime rate is similar to residential instability and low rates of home ownership in terms of its role in affecting voting outcomes. Here, we use a restricted sample because of concerns about the quality of the crime data for some counties. 10 Results pre- sented in column 1 show that the percentage of women not in the labor force has a positive and significant effect on opposition to same-sex marriage when the crime rate is set at its mean value and that the effect is significantly stronger in counties with higher crime rates. The same pattern is seen when examining interactions between occupational sex segregation and crime, as well as percent married with children and crime. In the latter case, however, the coefficient of the interaction term is not significant at the .05 level (p = .064). The coefficient for the interaction of the crime rate and percent same-sex households is also significant, indicating that the negative effect of same-sex households on opposition to same-sex marriage is stronger in counties with high crime rates. However, we find that the interaction of the VOTING TO BAN SAME-SEX MARRIAGE--905

DISCUSSION
For several decades, and especially in recent years, same-sex marriage has been a hotly contested public issue in the United States. The fight will certainly continue. Recent court rulings in California and New York have favored proponents of same-sex marriage, yet opponents continue to leverage public opinion by using ballot initiatives to ban same-sex marriage through amendments 906--AMERICAN SOCIOLOGICAL REVIEW   .190 .213 .197 .198 . to state constitutions. Although most of these electoral battles have been contested in states where voters are relatively conservative and tend to align with Republican candidates, same-sex marriage opponents have demonstrated that they can also win battles in Democratic strongholds such as California and Michigan. Holding statelevel differences constant, we sought to identify features of local settings that account for varying levels of support for and opposition to samesex marriage.
Counties characterized by traditional family structures and gender roles tend to show strong opposition to same-sex marriage. We also find that the effects of traditional family structure and gender roles are significantly stronger where community cohesion is weak, as indicated by residential instability, low rates of home ownership, and high crime rates. We explained these findings by arguing that traditional gender roles and family structures in local settings create a situation in which many residents are likely to view same-sex marriage as threatening to their interests and values. In such contexts, the prevalence of traditional arrangements means that individual perceptions of threat are likely to be continually reinforced through social interaction, as individuals interact on a regular basis with others who share similar views by virtue of their shared circumstances. These conditions, we argue, predispose community members to perceive same-sex marriage as a threat to the community, particularly in locations where community cohesion is already in peril.
In making these arguments, we drew on two different strands in the literature that aim to explain the more general issue of prejudice based on sexual orientation. One strain emphasizes how individuals who enact traditional gender roles have a vested interest in maintaining the status quo, as they view same-sex marriage and public acceptance of homosexuality as threatening to their own interests. The other strain emphasizes community characteristics and suggests that sexual-orientation prejudice reflects, in part, a reaction to social disorganization and weak community cohesion, which can generate feelings of distrust and intolerance of difference. Our research indicates that it is important to examine how these two general conditions (traditional gender roles and weak community cohesion) work together to produce varying levels of opposition to same-sex marriage. Our measures of residential instability and low home-ownership, by themselves, are each associated with lower levels of opposition to same-sex marriage, which should not be surprising. In local contexts where there is no broad consensus concerning the sanctity of the traditional family, the fight against same-sex marriage is likely to be viewed as an unnecessary distraction. Rather than perceiving same-sex marriage as a prime threat to the community, residents may be more inclined to see deficient schools, poverty, or discrimination as the most pressing community problems.
Although we did not focus on economic circumstances when making our theoretical argument, our analyses do show that opposition to same-sex marriage tends to be stronger in less prosperous counties. More specifically, holding other factors constant, opposition to samesex marriage tends to be weaker in counties with high median family incomes, higher levels of education, and higher percentages of the population enrolled in college and employed in professional occupations. Opposition tends to be stronger in counties with higher percentages of workers employed in production and construction occupations, and where higher percentages of households receive public assistance income (although the latter finding is not statistically significant).
These findings might be reconciled with a threat-based argument, such as the one we made above. Where there is economic deprivation and limited opportunities for social mobility, community residents are more likely to perceive themselves as vulnerable to economic competition that may result when discriminatory barriers are lifted (Bernstein et al. 2003;Persell et al. 2001;Stein 2001). Other researchers, however, might characterize these findings as an example of low-status backlash. Theories of status politics and status substitution, for example, propose that the demand for "moral reform" comes from individuals who feel compelled to assert the superiority of their own way of life as a way of compensating for their limited or declining economic power and influence (see Gusfield 1963;Lipset and Raab 1970). In this perspective, the negative orientation toward same-sex marriage has less to do with threats to privileges and advantages enjoyed by some heterosexuals, and more to do with resentment and frustration expressed by those who are not prospering economically.
Although our data do not allow us to settle this dispute definitively, we think that it is dangerous to assume that coercive moralism is the exclusive property of the economically disadvantaged. In her study of the late-1800s' antivice movement, for example, Beisel (1997:4) shows that the movement's success can be explained, in part, by the way that "its leaders connected concerns about the moral corruption of children, occasioned by changes in gender roles and the social meaning of sexuality, to threats to the social position of the upper and middle classes." In the same way, we expect that in local settings where traditional gender roles and family structures are accompanied by prosperity, the perception of threat may be heightened, rather than reduced, because individuals with a vested interest in maintaining traditional arrangements have a greater stake in the outcome.
To test this proposition, we estimated 30 separate models examining interactions between each of our five measures of traditionalism and six measures of prosperity. Rather than present the full results of 30 different models, we summarize the results pertaining to the interaction effects in Table 6. 12 The coefficient for the interaction term is statistically significant in all but 4 of our 30 models. The results indicate that although prosperity is related to less opposition to same-sex marriage, the positive effects of our measures of traditionalism on opposition to same-sex marriage are strongest in more prosperous counties. Looking at the cell in the topleft-hand corner of Table 6, for example, we see that a high percentage of women not in the labor force is significantly related to opposition to same-sex marriage when our measure of median income is set at its mean level, and that the effect is stronger in counties with higherthan-average median income. Figure 2 displays this same relation graphically. As the figure shows, opposition to same-sex marriage tends to be high in low-income counties regardless of the extent of women's participation in the labor force. Opposition is lowest in counties with high median incomes and high proportions of women participating in the paid labor force. Yet in counties where roughly 50 percent of women do not participate in the labor force, opposition to same-sex marriage is just as strong among high-and low-income counties.

CONCLUSIONS
A majority of U.S. citizens oppose same-sex marriage, and this opposition can be strong even in locations where residents tend to be prosperous and highly educated. Yet support and opposition vary substantially across local contexts. In this article, we account for that variation by considering how the organization of social life in local settings influences the extent to which community residents view same-sex marriage as threatening to both their own interests and the community at large. We focus on counties as units of analysis, rather than individuals, because perceptions of threat should depend to a large extent on the type of community in which an individual is embedded. In Martin County, Texas, for example, where 95 percent of voters approved of a ban on same-sex marriage, sex differences are established and reinforced through the organization of economic and family life. Nearly one half of adult women (46.7 percent) reported that they did not work in the paid labor force. Women who did work labored within a highly sex-segregated market. Based on our data, approximately 75 percent of women would have to switch occupational categories to produce an even distribution of women across all categories. The perceived reality in such a setting, regardless of individual attributes, is undoubtedly very different from that in Boulder County, Colorado, where only 33 percent of voters supported the ban. In Boulder, 75 percent of women work in the labor force and less than 40 percent of those women would have to change occupational categories to produce a perfectly integrated labor market.
Our structural approach in this article has its limits. Most importantly, in seeking to identify patterns of voting outcomes at the county level, we do not directly observe the micro-level processes that produce these patterns. However, we think that attempts to identify micro-level processes are facilitated by knowledge of macrolevel patterns. We hope that our work will stimulate more research at both the micro-and macro-levels, and we hope this research will help us better understand why a majority of Americans continue to feel justified in denying marriage rights to samesex couples.  (Table SF1 P20).
Percent Same-Sex Households: Percent of all households reporting a male householder and a male partner or a female householder with a female partner. Data taken from the 2000 U.S. Census (Table SF4 PCT21).
Percent Unmarried, Opposite-Sex Households: Percent of all households reporting an unmarried male householder and a female partner or an unmarried female householder and a male partner. Data taken from the 2000 U.S. Census (Table SF4 PCT 21).
Residential Instability: Percent of individuals age 5 years and older who are living in a different residence than the residence in which they lived five years earlier. Data obtained from the 2000 U.S. Census (Table SF3 P24).
Percent Homes Not Owner Occupied: Percent of housing units not occupied by the owner.
Crime Rate: Number of index crimes per every 1,000 county residents. Data taken from the 2000 Uniform Crime Report. We accessed the data through a Web site administered by the Inter-university Consortium for Political and Social Research (ICPSR). For a more detailed discussion of the crime measure, see the text of this article or McVeigh (2006).
Percent Production or Construction Occupations: Number of individuals employed in "construction, extraction, and maintenance" occupations or "production, transportation, and material moving" occupations, as a percent of the employed civilian population age 16 and over. The occupational categories are based on the Standard Occupational Classif ication Manual. Data taken from the 2000 U.S. Census (Table SF4 P86).
Percent Professional Occupations: Number of individuals employed in "professional and related occupations" as a percent of the employed civilian population age 16 and over. The occupational category is based on the Standard Occupational Classification Manual.  There is no space prior to