Religion, Marriage Markets, and Assortative Mating in the United States

Abstract

As interfaith marriage has become more common, religion is thought to be less important for sorting partners. However, prior studies on religious assortative mating use samples of prevailing marriages, which miss how local marriage markets shape both partner selection and marriage timing. Drawing on search theory and data from 8,699 young adults (ages 18–31) in the National Longitudinal Study of Youth 1997, the author examined the association between the concentration of co-religionists in local marriage markets and marriage timing and partner selection using event history methods. Religious concentration is associated with higher odds of transitioning to marriage and religious homogamy (conditional on marriage) for women and men at older ages (24–31) but not at younger ages (18–23). The association was also stronger for non-Hispanic Whites compared to other race-ethnic groups. The findings indicate that religion remains relevant in sorting partners for many young adults in today’s marriage market.

The rise of religious intermarriage is one of the most striking changes in partnering behavior in the United States (Kalmijn 1991; Rosenfeld 2008). Nearly a century ago, less than a third of all marriages were between partners of different religious upbringings and only 10% of couples maintained separate religious identities after marriage. Today, over half of all marriages cross religious lines, with some 40% of couples remaining interfaith (Putnam & Campbell 2010). Attitudes towards religious intermarriage among young adults and their parents have also become more accepting (Putnam & Campbell 2010). These trends, coupled with more general declines in religious identification and attendance (Hout & Fischer 2014; Pew Research Center 2015; Schwadel 2010), have led scholars to pay less attention to religion in partnering decisions.

Yet, despite a decline in its importance, there are signs that religion remains relevant in the marriage market. First, religious similarity between partners seems to matter more to some religious groups than others: compared to Mainline Protestants and Catholics, Evangelical Protestants and Mormons have maintained relatively high levels of religious homogamy (Lehrer 1998; Logan et al. 2008; Sherkat 2004). Second and more generally, cultural compatibility remains an important basis on which people choose romantic partners (Kalmijn 1998; Schwartz 2013). As a signal of a person’s beliefs and practices, religion may continue to act as a filter for many men and women seeking partners with similar values and cultural orientations—not just for the religiously devout but also for atheists and the religiously unaffiliated.

This study revisits religion’s role in sorting partners using a marriage-market approach. Most prior research examines national trends in religious homogamy among prevailing marriages to assess whether religion has become more or less important for marriage decisions over time. In contrast, I draw on search theory (Oppenheimer 1988) to test how variation in the supply of religiously similar partners across local marriage markets is related not only to partner choice but also to entry into marriage, or marriage timing. I ask whether young adults in unfavorable markets with few religious matches delay marriage rather than intermarry and whether those in favorable markets find a religious match more easily and marry more quickly.

I use data from the first fifteen rounds of the National Longitudinal Survey of Youth 1997 (NLSY97) and event history methods to model the associations between the local concentration of co-religionists and marriage timing and partner’s religion (homogamy vs. heterogamy) among young adults ages 18–31. I also test whether the association between religious concentration and marriage varies by religious tradition and religious commitment. In keeping with prior studies on marriage, I analyze respondents separately by gender. I also incorporate recent changes in family formation by examining variation in marriage market effects at different stages in the young-adult life course as well as differences between entry into marriage and cohabitation.

Background

Religious Assortative Mating in the United States

Religion was once an important social boundary in the marriage market. Few Americans married a spouse who did not share their religion or denominational affiliation, and those who did often switched religions upon marrying (Kalmijn 1991; Kennedy 1944; Glenn 1982; Lehrer 1998). Religious boundaries were reinforced, in part, by strong prohibitions by families and religious communities against exogamy. In addition, shared religious beliefs and practices were thought to provide spouses with a strong foundation on which build a happy and fulfilling marriage.

Today, over half of all marriages are between partners of different religious upbringings and 40% of couples maintain separate religious identities after marriage (Putnam & Campbell 2010). As religious homogamy has declined, sorting on educational attainment and income has increased (Schwartz & Mare 2005; Schwartz 2010). Some scholars attribute these changes to shifts in partner preferences, arguing that “achieved” traits (education and income) have supplanted “ascribed” traits related to family background (religion, race, class) as the primary basis for sorting partners (Schwartz 2013). Indeed, today’s young adults are more likely than their predecessors to approve of religious intermarriage (Putnam and Campbell 2010). Changing preferences are likely reinforced by growing religious diversity, declining religious authority and membership (Chaves 1994; Hout & Fischer 2014), decreasing salience of denominational boundaries (Wuthnow 1988), socioeconomic convergence between religious groups (Kalmijn 1991), and weakening third-party control over mate selection (Rosenfeld and Kim 2005).

While trends over time suggest that religion has become less important for partnering decisions, there are reasons to think that it remains relevant in the marriage market. Churches and religious organizations remain important centers of social life for many Americans, particularly during childhood (Feld 1989; Smith and Denton 2005). Congregations often facilitate partner selection by providing young adults with opportunities to meet partners of similar religious backgrounds (Kalmijn & Flap 2001). Most religions also actively encourage homogamy to help grow and retain members (Sherkat 2004; Hout et al. 2001; Voas 2003)

Religious similarity may also be attractive as a signal of cultural compatibility. In marriage, as in friendships, people are attracted to others who share their values and lifestyle (McPherson et al. 2001; Kalmijn 1994). Religions constitute an important social source of beliefs and attitudes towards issues that matter to potential spouses: family formation and parenting, priorities around family and work, division of labor in the household, politics, and identity (Smith 2003; Mahoney 2010). Religions also involve practices and rituals that bind people to communities and provide connection to the past (Tweed 2008). Even for atheist and non-religious young adults, commitment to secular values may be an important filter during the search for a spouse.

Search Theory and Marriage Markets

However, assortative mating trends are limited in their ability to reveal religion’s role in sorting partners. By focusing on prevailing marriages, often at the national level, trend studies miss the dynamics of partner selection and how the local availability of particular kinds of partners influences who people marry and when they marry.

I draw on an alternative approach based on Oppenheimer’s (1988) search theory of marriage timing to understand the importance of religion for marriage decisions. Search theory relates marriage timing to how easily people can find and secure a spouse who matches their preferences (Lewis & Oppenheimer 2000; Raley & Bratter 2004). Partner preferences are multidimensional, encompassing subjective qualities as well as social characteristics, such as socioeconomic status, race-ethnicity, and religion (Kalmijn 1998). Certain traits can be valued more highly than others and take greater precedence in decision-making. For instance, some young adults may be willing to marry outside their own race-ethnic group if it means marrying a spouse with good economic prospects. Men and women’s own attractiveness (physical, economic, and social) also affects their marriage timing through assortative mating. Highly-desirable adults can attract and find a spouse who matches their preferences more easily, which also allows them to enter marriage quickly.

Assortative mating—and, therefore, marriage timing—is also sensitive to the availability of partners in the local marriage market. In favorable markets, where suitable partners are abundant, finding a spouse who matches one’s preferences is relatively easy, increasing assortative mating on a particular characteristic (Blau et al. 1982) while also hastening the transition to marriage. Likewise, finding a well-matched partner in unfavorable markets is much harder. If a given trait is important, young adults may be unwilling to intermarry, instead opting to delay marriage and continue searching for a partner. But there are costs to continued search: financial and emotional costs to dating, opportunity costs to turning down partners as the market thins out, as well as social costs. In communities where marriage is highly valued, young adults may choose to intermarry rather than remain single. In this way, market variation in partner supply and marriage behavior can reveal the strength and salience of a given partner trait.

Consistent with search theory, prior research finds that marriage timing and assortative mating are sensitive to the demographic availability and socioeconomic attractiveness of potential partners in the local market (Lichter et al. 1992; Lichter et al. 1995; Lewis & Oppenheimer 2000; Guzzo 2006). Only one study has examined the connection between the religious composition of the marriage market and religious assortative mating (Lehrer 1998). As expected, married couples were more likely to match on religion in areas with higher concentrations of same-faith partners. However, this study only looked at prevailing marriages, ignoring marriage timing. In this study, I apply search theory to religious assortative mating to ask: When few religious matches are available in the local market, do young adults delay marriage or compromise and intermarry? Likewise, in markets with lots of religiously similar partners, are young adults more likely to find a partner and enter marriage?

Hypotheses

I expect that religious similarity is relevant to young adults when choosing a spouse, even if religion’s influence is weaker now than in previous eras. Religion should not only be a reason to accept or reject a partner, but religious similarity and shared cultural background should also strengthen budding relationships and hasten the transition to marriage. I expect the local supply of partners to be positively associated with marriage timing and partner selection because favorable markets will enable young adults to find a religious match more easily.

Hypothesis 1: Young adults in markets with more co-religionists will enter marriage more quickly, net of other factors, compared to young adults in less favorable markets.

Hypothesis 2: Upon marriage, young adults in markets with more co-religionists will be more likely to marry a same-faith partner compared to those in less favorable markets.

Importantly, Hypothesis 2 could find support even if religion is irrelevant to young adults’ marriage decisions. Population composition is an important structural determinant of intergroup relations (Blau et al. 1982). The chances of a religious match, given the decision to marry, will necessarily depend on the religious composition of the market. What is crucial for discerning religion’s relevance for marriage decisions is the association between religious market concentration and marriage timing—any association with partner choice only provides confirmation of search theory’s expectations.

Religious market concentration may also matter more for women than men. Women in the U.S. tend to be more religious than men (Pew Research 2016) and are viewed by many religious traditions as the parent responsible for religious socialization of children (Glenn 1982). Women also tend to be closer with their family-of-origin (Lee et al. 2003), allowing parents to more effectively pressure daughters to marry within their religion than sons.

Hypothesis 3: The association between religious concentration and marriage behavior will be stronger for women than men.

There may also be variation by religious tradition. Although intermarriage has increased for all faiths, change has been less pronounced among conservative Protestant traditions (e.g., Evangelicals, Mormons) (Lehrer 1998; Sherkat 2004). These differences in intermarriage rates could reflect partner preferences. Among never-married young adults, ages 18–23, surveyed in the 2008 National Study of Youth and Religion, some 50% of Mormon youth and 39% of Evangelicals reported that marrying a same-faith partner was very or extremely important to them, compared to just 19% of Mainline Protestant youth and 13% of Catholics and Jews (author’s calculations). This suggests that the significance of partner’s religion for marriage decisions will be greater for young adults from conservative religious traditions. They should be more responsive to local market conditions by entering a religiously homogamous marriage more quickly when partners are numerous and delaying marriage when partners are in short supply.

Hypothesis 4: The association between religious concentration and marriage behavior will be stronger for Evangelical Protestant and Mormon young adults, net of other factors, compared to young adults from more liberal traditions and the non-religious.

Alternatively, there may be greater variation by religious commitment than religious tradition. Religious participation declines more dramatically during young adulthood than religious affiliation (Uecker et al. 2007), leaving a large group of nominally religious young adults with weak attachment to the religion in which they were raised. There also tend to be differences in marriage timing and other family behaviors between religiously committed and un-committed young adults, net of religious affiliation (Lehrer 2004; Uecker & Stokes 2008; Uecker 2014). Young adults with greater attachment to and participation in their religion may value religious similarity with their spouse more highly compared to less religious young adults.

Hypothesis 5: The association between religious concentration and marriage behavior will be stronger for highly religious young adults, net of other factors, compared to less religious young adults.

The importance of religion for partnering decisions is also likely to vary by life course stage. In recent decades, the average ages at first marriage for men and women have climbed to 27 and 29 (Census 2011). Most young adults are not on the marriage market until their mid to late twenties, after they have completed their education and started careers (Cherlin 2004). Those who do marry at early ages tend to be more religiously devout and affiliated with religious traditions that prioritize family roles and discourage premarital sex and cohabitation (Glass and Jacobs 2005; Uecker & Stokes 2008; Uecker 2014). For early-marriage youth, religious homogamy is likely to be important—but there are also likely to be higher social costs associated with remaining unmarried relative to marrying outside their religious denomination. Moreover, young adults wishing to marry earlier than their peers may also search for partners within socially circumscribed religious communities, making marriage decisions less dependent on the composition of the wider market. In contrast, young adults searching for spouses in their late-twenties may be influenced more heavily by the composition of the marriage market. Not only are more people searching for a spouse at these ages, making marriage more likely when a match is found, but the market has yet to thin out, keeping the opportunity costs of continued search relatively low compared to later ages.

Hypothesis 6: The association between religious concentration and marriage behavior will be stronger for young adults during the mid- to late-twenties, net of other factors, compared to earlier ages.

Finally, I also consider how religious composition of the market might influence entry into cohabitation. Cohabitation is increasingly common in the United States, accounting for 73% of first unions in 2001–2007 (Kennedy & Bumpass 2011). Still, evidence suggests that marriage remains distinct from cohabitation (Smock 2000; Sweeney 2010). Most cohabitating unions are short-lived and only half lead to marriage (Kennedy and Bumpass 2008). This suggests that religious similarity between partners may be less important when young adults are contemplating cohabitation compared to a life-long commitment in marriage (Blackwell & Lichter 2004). In addition, marriage-market studies looking at the sex ratio and the economic characteristics of men find that women’s transition into cohabitation is less sensitive to local conditions than the transition into marriage (Raley 1996; Guzzo 2006).

Hypothesis 7: The local religious concentration should be more strongly associated with the transition to first marriage than to first cohabitation.

Method

Data

Data for the analysis come from Rounds 1–15 of the 1997 National Longitudinal Survey of Youth (NLSY97), a national sample survey of 8,984 youth, born between January 1980 and December 1984. Respondents have been surveyed annually since 1997 and asked, among other topics, about their family formation histories. With the use of the sampling weights provided, the NLSY97 is designed to be nationally representative. I used sampling weights from the first round for all analyses. The restricted version of the data provides identifiers of the county in which respondents reside at the time of each interview. This information allowed me to merge time-varying data from the census and other sources onto the individual-level data file to describe local marriage market conditions. Data used in this analysis cover years up to 2011 (Round 15) when respondents were 26 to 31 years old. As a result, this study reflects relatively early marriage for today’s young adults, many of whom will marry in their thirties (Census 2011). Still, given the wealth of detailed data on respondents’ religion, partner characteristics, place of residence, as well as education and work histories, the NLSY97 is an ideal dataset for testing hypotheses about the relationship between marriage markets and marriage behavior.

I converted the data into person-years using birthdates and dates of first marriage. This person-year data set included up to 13 observations per individual starting with each respondent’s 18th birthday, including all person-years up to the year of first marriage or last interview. Respondents who had missing data on the date of first marriage (N=66), who married prior to age 18 (N=75), or whose last interview as of round 15 was prior to age 18 were excluded from the sample (N=144). This left a sample of 4,216 women and 4,483 men. An addition 460 first-married respondents were missing data on partner’s religion. To predict the transition to marriage, I used respondents’ characteristics at their birthday month in one year (e.g. age 24) to predict their probability of marriage by the following year (e.g. age 25).

Measures

Dependent Variables

The first outcome was a dichotomous variable measuring the transition into first marriage, equal to 0 prior to first marriage and equal to 1 in the first year of marriage. In the sample (N=8,699), 1,850 women and 1,589 men were married as of their last interview. Annual probabilities of marriage for women and men in this cohort during this period in the NLSY97 sample are comparable to those found in the 2011 American Community Survey. The second outcome was a three-category variable measuring the transition into first marriage based on respondent and partner’s religious affiliations, equal to 0 prior to first marriage, 1 in year of first marriage for religiously homogamous unions, and 2 for religiously heterogamous unions. The third outcome variable was a three-category variable measuring the transition into cohabitation or marriage as first union type, as opposed to remaining unpartnered.

To measure the second outcome, I used respondent’s self-reported religious affiliation at Round 1. NLSY97 distinguishes between 25 different religious affiliations. Using the self-report measure, I collapsed categories to approximate the RELTRAD classification (Evangelical Protestant, Mainline Protestant, Catholic, Jewish, Mormon, Unaffiliated, Other [Steensland et al. 2000]). Partner’s religious affiliation is asked of respondents in an identical manner as part of the household roster questionnaire at rounds 2–7 and again at round 15. I used the first report of partner’s religious affiliation because spouses tend to become similar on religious affiliation the longer they are together. At each interview after round 7, NLSY97 asked respondents who either had (1) married but were not interviewed during rounds 2–7, or (2) married after round 7 but divorced before round 15 to report their partner’s religious affiliation when they first married.

This method posed challenges for measuring religious homogamy at the time of marriage. First, because respondents were only asked to update their religious affiliation periodically across survey rounds, I measured religious affiliation as a fixed characteristic from the first interview. Although young adulthood is characterized by identity formation and change, this choice may be less problematic because changes in religious affiliation are less common than declines in religious participation (Uecker et al. 2007). Still, because respondents may have switched religions prior to meeting their partner or converted to marriage to match their spouse, my method likely underestimates the prevalence of religious homogamy. Additionally, some respondents reported their partner’s religious affiliation years after they first met or married, which could also lead to a mis-estimation of homogamy if partner’s religion changed between time of marriage and time of measurement (e.g., because the partner converted to respondent’s religion). However, as discussed previously, the analysis of assortative mating is less central than the analysis of marriage timing, which is unaffected by measurement of religious homogamy.

Among married female respondents with data on partner’s religion, 855 were married to spouses of the same religion and 782 to spouses of a different religion. Corresponding figures for married male respondents were 675 and 667. Table A1 in the online appendix shows religious assortative mating patterns among married respondents and their spouses by religious tradition. Consistent with prior research on religious intermarriage and stated partner preferences on surveys (Lehrer 1998; Sherkat 2004), Evangelical Protestant, Jewish, and Mormon young adults were more likely to marry a spouse of the same religion compared to other groups.

Marriage Market and Contextual Characteristics

I defined marriage-market boundaries by commuting zones (Autor & Dorn 2013). These are clusters of counties with strong commuting ties (defined by the Census) and are the best available approximation to “labor market areas” (LMAs) used in previous marriage-market studies from the 1990s (e.g. Lichter et al. 1992). The 741 commuting zones cover the entire United States and can cross state boundaries. The link between counties and commuting zones provides stability to market boundaries over time and allowed me to track NLSY-97 respondents across survey rounds. In addition, they represent socially meaningful local boundaries (areas with strong work commuting ties) in contrast to politically meaningful boundaries such as states or counties. All local characteristics were calculated at the commuting zone.

To calculate the focal independent variable, religious concentration of the local marriage market, I used data from the 2000 and 2010 Religious Congregations and Membership Study (RCMS) administered by the Association of Statisticians of American Religious Bodies (Jones et al. 2000; Grammich et al. 2010). In 2000 and 2010, researchers reached out to some 250 religious bodies and asked them report the numbers of their congregations and religious adherents in every county. After collecting the data, researchers checked for accuracy and consistency and, in some cases, adjusted numbers based on previous religious censuses and other data sources (see www.thearda.com for more details). The RCMS is the best and only available data source with time-varying estimates of the size and share of detailed religious denominations at the county-level. The RCMS also has several shortcomings, which I discuss below.

Using the RCMS, I classified religious traditions according to the RELTRAD method and distinguished between Evangelical Protestants, Mainline Protestants, Catholics, Jews, Mormons, and other religious traditions (e.g. Muslims, Hindus, Buddhist, Jehovah’s Witnesses). “Other” religious groups are a small proportion of the US population and make up just 2% of respondents in NLSY97. I aggregated county estimates of the number of religious adherents in each group up to the commuting zone. Because RCMS does not have estimates of the numbers of religiously unaffiliated, I used the Pew Research Center’s 2007 and 2014 Religious Landscape surveys to measure the percentage of the population with no religious affiliation. Although both surveys are large (~35,000 respondents each), estimates of the unaffiliated could only be made at the state-level because county of residence was not available. Thus, prior to aggregating the unaffiliated numbers up to the commuting zone level, each county was given its state’s value.

One shortcoming of the RCMS is that some groups chose not to participate in the study, which led to undercounts of religious adherents in many counties. In addition, the number of religious adherents in a given county was calculated based on church attendance and membership, not place of residence of congregants. As a result, the number of religious adherents can exceed the number of people in counties where large numbers of people from neighboring counties cross county lines to attend services (Finke & Scheitle 2005). This issue is less problematic for commuting zones (counties with strong commuting ties). Still, to account for both issues, I adjusted the religious composition of each commuting zone to simulate complete coverage, or 100% of the population (similar to Glass & Levchak 2014).

I defined religious concentration as the percentage of the commuting-zone population with the same religious tradition as respondents. For example, for Catholic youth, religious concentration is the percentage of Catholics in commuting zone of residence. Because some religious traditions are better represented across all commuting zones, I used a relative measure of market concentration rather than an absolute measure. This is a group-standardized variable measuring the degree of religious concentration in a given respondent’s market relative to the average market faced by members of that religious tradition. I calculated means and standard deviations of religious concentration for each religion and expressed a given respondent’s score in standard deviation units from the mean for those of the same religion. This approach better captures the effect of market variation in religious concentration on marriage behavior. For example, take two fictional marriage markets, Newoak and Wellston. Newoak is 10% Jewish and 90% Catholic, while Wellston is 30% Jewish and 70% Catholic. In absolute terms, both towns are more favorable markets for Catholic youth compared to Jewish youth. But in relative terms, Wellston is a more favorable market for Jewish youth, whose peers in Newoak have fewer religiously similar partners, while Newoak is a better market than Wellston for Catholics.

The RCMS data has two other shortcomings. First, RCMS does not have demographic information about religious adherents, which makes it impossible to calculate age-, gender-, or race-specific measures of the local religious composition. Lack of information on race-ethnicity may be particularly problematic given that race is a strong division in the marriage market (Rosenfeld 2008) and religious congregations are heavily segregated by race-ethnicity (Emerson & Smith 2001). Moreover, because non-Hispanic whites are the largest group in most commuting zones, RCMS data reflects the religious composition for whites much better than for minorities. Second, RCMS has poor coverage of Black Protestant denominations. Given that the majority of African Americans self-identify with a Black Protestant denomination (53% in 2014 [Pew Research Center 2015]) and Black Protestant congregations are concentrated in the South, this could lead to biased estimates of the religious composition of some counties, particularly those with large black populations (Finke & Scheitle 2005).

I addressed these shortcomings in two ways. First, in addition to main models that include all respondents, I estimated separate models for non-Hispanic whites and non-white respondents. This helped account for better measurement of market characteristics for whites compared to other groups. I expected that religious market concentration would be more strongly associated with marriage behavior for non-Hispanics whites than for other groups. Second, I created an alternate measure of religious concentration that adjusts for the race-ethnic composition of commuting zones and the race-specific religious composition of states. I used data from Pew Research Center’s 2007 and 2014 Religious Landscape Surveys to estimate the religious composition of states separately by race-ethnicity. I then calculated a weighted average between RCMS composition of commuting zones and Pew’s race-specific composition of states for each race-ethnic group based on their share of the population in each commuting zone. This procedure gave greater weight to the RCMS commuting zone estimates for a race-ethnic group when that group was a larger share of the population and greater weight to state estimates when the group was a smaller share of the population.

I controlled on other characteristics of the marriage market associated with marriage behavior using 2000 Census 5% sample and the 2006–2010 American Community Survey 5-year sample, linearly interpolating across years (Ruggles et al. 2015). The sex ratio, or sex composition, is the sex- and race-specific percentage of unmarried opposite-sex partners in the non-institutionalized population ages 18–34 (% female for males; % male for females). This captures demographic availability and competition for partners in the local market (Fossett & Kiecolt 1991; South et al. 2001). I measured socioeconomic characteristics of partners (by race and sex) in two ways. Educational concentration is the share of opposite-sex partners with a bachelor’s degree. Economic attractiveness is measured as the percentage of opposite-sex partners who are employed full-time (>35 hours per week).

Differences in religious concentration could also be associated with the level of religious diversity in the market as well as broader attitudes towards family in the community. For instance, markets with greater religious diversity—but lower concentrations of any one religious group—might promote homogamy by making religious identities and boundaries more salient to potential partners. Likewise, areas with large concentrations of religious and political conservatives might influence all young adults’ marriage timing, regardless of the concentration of co-religionists (Glass & Levchak 2014). To measure religious diversity, I used RCMS data to calculate a modified Herfindahl-Hirschman Index, with higher scores indicating greater religious diversity (Pew Research Center 2014). To measure cultural conservatism, I used the percent of the commuting-zone population that voted for George W. Bush in the 2004 presidential election—an election in which cultural attitudes toward family issues and behaviors were salient and spatially concentrated (Lesthaeghe & Neidert 2006).

I controlled for the annual unemployment rate using data from the Bureau of Labor Statistics to capture the broader economic climate of respondents’ surrounding area related to employment prospects and economic security (Harknett & Kuperberg 2011). The NLSY-97 cohort experienced the 2001 and the 2007–09 recessions. Using the 2000 and 2010 censuses, I controlled for socio-demographic characteristics of the local area that are likely to be related to marriage behavior: age structure (25–44, 45–64, 65+), total population size (logged), education (% of 25+ population with BA or more) and race-ethnic composition (%’s Black and Hispanic).

Individual Characteristics

I also controlled for respondent characteristics relevant to marriage behavior. In addition to religious affiliation, respondent religiosity was measured by weekly attendance at religious services (=1 weekly attendance or more). Parental religiosity was a composite scale of items, asked at round 1, measuring religious beliefs, salience, and practices (range 1–6). Educational attainment was measured as a time-varying monthly indicator that distinguishes between a college degree or more, some college (associate’s degree), a high school diploma, and less than high school. I controlled for school enrollment using a monthly dichotomous indicator (1=enrolled). To measure young adults’ earnings, I used annual self-reports on the total income received from wages, salary, commissions, or tips from all jobs in the prior year. I created measures of employment hours based on a NLSY97-created variable describing number of work hours each week. I averaged weekly counts to create monthly indicators, recoded into 3 categories: not employed (looking for work or out of labor force), part-time employment (< 35 hours a week), and full-time (35 hours a week or more).

I also controlled for demographic and family background characteristics that are known predictors of marriage including family structure at age 12 (two-biological parents, single mother, stepparent, and other) and parental education (whether respondent’s most highly educated parent had a college degree). Because premarital childbearing can influence parents’ subsequent opportunities to marry (Bennett et al. 1995), I controlled for respondents’ first birth status using a four-category time-varying variable that takes into account children’s life stages (Manning 1993). I distinguished between person-years without a child, gestational period (7 months to month prior to first birth), year 1, and all subsequent years. I also controlled for region and metropolitan residence because young adults in the South and rural areas are more likely to marry at earlier ages and marry assortatively on religion (Sherkat 2004; Uecker 2014).

In models for older young adults, ages 24–31, I also include a sample selection correction using a two-stage procedure described by Heckman (1979) to account for potential bias in the estimates resulting from the exclusion of respondents who married (or cohabited in the first union analysis) prior to age 24. First, I constructed a probit model to predict selection into the sample using race-ethnicity, family structure at age 12, and parental education for men and women separately. A hazard rate for exclusion from the sample was then constructed based on the predicted values from this model (see Berk 1983). The hazard rate was entered as a control into the equations predicting marriage timing, spouse’s religion, and first union type.

To deal with missing data, I employed the multiple imputation technique in STATA 13, using the Chained equations in MI command (White et al. 2011). I generated five imputed data sets to impute missing data on all categorical and continuous independent variables included in the analysis (see Tables 1 and 2). Five imputations are considered sufficient to make good inferences and to get parameter estimates that are close to being fully efficient (Allison 2009). Results were substantively equivalent before and after multiple imputation. I present results from models using the multiply imputed data.

Table 1.

Sample fixed characteristics (weighted %)

	Analytic sample		Analytic sample (based on multiple imputed person-year data)
	Women	Men	Women	Men
No. of respondents	4,216	4,483
No. of person-years	35,219	40,158
No. of first marriage by age 23	1,052	752
% religiously homogamous	53.2	53.2
% religiously heterogamous	46.8	46.8
No. of first marriages by age 31	1,850	1,589
% religiously homogamous	52.5	50.3
% religiously heterogamous	47.5	49.7
No. of first unions	2,823	2,809
Religious affiliation in 1997
Evangelical Protestant	38.4	34.3	37.3	33.9
Mainline Protestant	19.8	17.8	19.6	17.0
Catholic	24.9	28.6	25.9	29.2
Jewish	1.4	1.2	1.5	1.3
Mormon	1.6	1.3	1.4	1.1
None	11.9	14.4	12.1	14.8
Other	2.0	2.4	2.2	2.6
Family structure at age 12
Two biological-parents	48.9	51.3	48.1	50.6
Single mother	34.6	32.3	35.3	33.3
Step-parent family	7.2	5.6	7.1	5.5
Other family types	9.3	10.7	9.6	10.6
Parental educational attainment
Less than high school	11.5	12.6	11.8	13.0
High school	31.2	30.3	32.4	31.8
Some college	25.1	24.3	26.2	25.0
College or more	27.5	28.3	29.5	30.2
Missing	4.7	4.5

Note: 460 first-married respondents (213 women, 247 men) had missing data on partner’s religion.

Table 2.

Individual time-varying characteristics and marriage market characteristics for women and men by life course stage, weighted, multiply-imputed person-year data

	Women				Men
	Ages 18–23		Ages 24–31		Ages 18–23		Ages 24–31
	Mean	SE	Mean	SE	Mean	SE	Mean	SE
Individual characteristics
Church attendance (1=weekly)	18.9		14.5		15.9		10.8
Parental Religiosity (1–6)	3.7	0.01	3.7	0.02	3.7	0.01	3.6	0.01
Educational attainment
Less than high school	23.9		9.1		28.3		10.9
High school	69.8		51.2		67.6		61.9
Some college	2.0		6.3		1.8		5.4
College or more	4.3		33.4		2.3		21.8
Employment status
Not employed	27.9		20.6		28.7		19.4
Part-time	48.1		33.4		34.6		23.7
Full-time	23.9		46.0		36.7		56.9
Enrolled in school	51.1		17.9		42.2		12.4
Annual earnings (logged dollars)	5.8	0.04	7.8	0.04	6.0	0.03	8.0	0.03
Marriage market characteristics
Religious concentration (% co-religionists)
Evangelical Protestants	34.8	0.24	34.1	0.27	34.6	0.23	35.1	0.26
Mainline Protestants	20.8	0.18	17.2	0.22	20.0	0.18	16.5	0.21
Catholics	42.2	0.21	40.2	0.25	41.9	0.19	39.2	0.22
Jews	2.4	0.04	1.5	0.03	2.3	0.03	1.3	0.03
Mormons	6.9	1.12	11.2	2.40	5.5	0.91	10.7	2.26
Religiously Unaffiliated	16.8	0.06	20.5	0.07	16.8	0.06	20.7	0.07
Other religious groups	4.2	0.21	4.2	0.20	4.5	0.16	3.3	0.14
Sex composition (% opposite-sex partners)	52.3	0.03	51.5	0.03	47.5	0.03	48.1	0.03
Educational concentration (% opposite-sex partners with BA or more)	14.7	0.07	16.1	0.09	18.0	0.07	19.9	0.09
Economic attractiveness (% opposite-sex partners employed full-time)	51.9	0.06	48.6	0.07	44.9	0.05	43.0	0.06
Religious diversity index	6.9	0.00	6.9	0.01	6.9	0.00	6.9	0.01
Cultural conservatism (% vote for Bush 2004)	51.8	0.08	50.3	0.10	51.3	0.08	50.6	0.09
Annual unemployment rate	5.1	0.01	6.9	0.02	5.1	0.01	6.9	0.02
Urban (%)	76.7	0.15	81.2	0.17	76.9	0.14	79.7	0.16
Non-Hispanic Black (%)	12.7	0.08	13.8	0.10	12.7	0.07	13.2	0.09
Hispanic (%)	11.6	0.09	14.5	0.12	11.7	0.08	14.0	0.11
Ages 25–44 (%)	0.3	0.00	0.3	0.00	0.3	0.00	0.3	0.00
Ages 45–64 (%)	0.2	0.00	0.3	0.00	0.2	0.00	0.3	0.00
Ages 65+ (%)	0.1	0.00	0.1	0.00	0.1	0.00	0.1	0.00
Total population size (logged)	14.0	0.01	14.3	0.01	14.0	0.01	14.2	0.01

Note: Religious concentration is group-standardized in regression models as described in the text. There is no missing data on market characteristics. If respondent's location was not available in a given survey round, respondents were coded as living in their last known residence.

Analytic Strategy

I employed logistic and multinomial regression to estimate discrete-time event history models, where time was measured by a series of dummy variables indicating respondent age, allowing the risk of marriage to vary non-parametrically (Allison 1982). I adjusted standard errors in all models to account for clustering of respondents in marriage markets using the Stata survey command –cluster–. First, I employed logistic regression to estimate discrete-time event history models of the transition to first marriage (marriage timing) that included religious market concentration as well as respondents’ characteristics and other characteristics of the marriage market (Hyp. 1). Next, to examine the effect of religious concentration on partner selection, I employed multinomial logistic regression to estimate competing-risk models that distinguished between marriage to a same-faith and different-faith spouse (Hyp. 2).

I ran separate models for men and women (Hyp. 3) and life-course stage (18–23/24–31; Hyp. 6). Models for ages 24–31 included a selection correction described above. Although splitting the sample at ages 23/24 is somewhat arbitrary, this approach roughly delineates between the years before and after school completion, when more young adults are entering the labor and marriage markets, and is consistent with previous studies using NLSY97 (Kuo and Raley 2014; McClendon et al. 2014). I conducted detailed analysis of how the association between religious concentration and marriage varied by age and observed a noticeable change in the early 20s which was clearest at ages 23/24. The results are substantively similar with other age breaks. To test whether the relationship between religious concentration and marriage varied by religious affiliation and religiosity (Hyp. 4 and 5), I ran models with interaction terms between religious market concentration and religious tradition and attendance. Finally, I estimated competing risk models to identify whether religious concentration was associated more strongly with first marriage than with first cohabitation (Hyp. 7).

Results

Marriage Timing

Table 3 shows the results from discrete-time event history models of the transition to first marriage for women, with separate models for early adults (ages 18–23) and older young adults (ages 24–31). In both age groups, non-Hispanic white women were more likely than other race-ethnic groups to enter first marriage. School enrollment was negatively associated with marriage for younger women while educational attainment and earnings were positively associated with marriage for older women. Net of socioeconomic characteristics and contextual controls, Mainline Protestants and Catholics were less likely to enter first marriage compared to Evangelical Protestants at younger ages but there were not significant religious affiliation differences between women at older ages. Weekly religious attendance was positively and strongly associated with first marriage across the young-adult life course. In contrast, parental religiosity was not associated with first marriage, net of controls. Living in the Northeast and in a metropolitan area was negatively associated with marriage for younger women, while pregnancy/having children were positively associated with marriage (not shown); these were not significant for older women. Other respondent characteristics were not statistically significant.

Table 3.

Coefficients from event history logistic regression models of transition into first marriage for women by life course stage (ages 18–23 and 24–31), weighted

	Ages 18–23			Ages 24–31
	B	SE B	OR	B	SE B	OR
Religious market concentration	0.01	0.04	1.01	0.09*	0.04	1.10
Religious affiliation (ref: Evangelicals)
Mainline Protestant	−0.24*	0.11	0.79	0.18	0.11	1.20
Catholic	−0.26*	0.11	0.77	−0.08	0.12	0.92
Jewish	−0.23	0.36	0.79	−0.24	0.34	0.79
Mormon	0.17	0.23	1.18	0.16	0.25	1.17
Unaffiliated	−0.08	0.12	0.92	−0.14	0.16	0.87
Other	−0.76†	0.45	0.47	−0.73†	0.43	0.48
Attends services weekly	0.51***	0.09	1.67	0.46***	0.10	1.58
Parental religiosity index	0.04	0.03	1.04	−0.01	0.03	0.99
Race-ethnicity (ref: NH White)
NH Black	−1.66***	0.25	0.19	−2.72**	0.87	0.07
Hispanic	−0.51*	0.22	0.60	−0.76***	0.22	0.47
Other	−0.80***	0.22	0.45	−1.19*	0.56	0.30
Educational attainment (ref: high sch.)
Less than high school	−0.36**	0.11	0.70	−0.17	0.18	0.84
Associate’s degree	0.52†	0.31	1.68	0.35*	0.17	1.41
Bachelor’s degree or more	0.23	0.19	1.26	0.48***	0.11	1.62
Enrolled in school (=1)	−0.50***	0.10	0.61	−0.05	0.12	0.95
Employment status (ref: not employed)
Part-time	−0.06	0.09	0.95	−0.13	0.13	0.88
Full-time	0.17†	0.10	1.19	−0.02	0.12	0.98
Annual earnings	0.03*	0.01	1.03	0.05**	0.02	1.06
Other market characteristics
Sex composition (% opposite-sex)	0.02	0.01	1.02	−0.02	0.02	0.98
Educational concentration	−0.02†	0.01	0.98	−0.01†	0.01	0.99
Economic attractiveness	−0.02**	0.01	0.98	−0.00	0.01	1.00
Religious diversity index	−0.10	0.09	0.90	−0.03	0.08	0.98
Cultural conservatism (% vote for Bush 2004)	0.01	0.01	1.01	0.01	0.01	1.01
Constant	−4.49*	1.91		−2.23	2.57
Person-years	21703			13789

Note: All models control for age, year, parent’s educational attainment, family structure at age 12, first birth status, metropolitan and regional residence, local age structure, race-ethnic composition, educational composition, annual unemployment rate, % urban, and total population size (logged). Standard errors are adjusted for clustering of respondents within commuting zones using Stata’s <cluster> command.

^†p<0.1

^*p<0.05

^**p<0.01

^***p<0.001

The focal association between religious concentration and first marriage timing for women varied across the life course. For younger women, there was not a significant association between religious concentration and first marriage. However, at older ages, women in markets with higher concentrations of religiously similar partners were significantly more likely to enter first marriage. In pooled models, the interaction between religious concentration and life course stage was statistically significant. The coefficient indicates that, relative to a respondent’s religious peers, residing in a marriage market with one standard deviation greater concentration of co-religionists was associated with a 10 % increase in the annual risk of marriage (exp[0.09] = 1.10). It is also instructive to consider the results in terms of unfavorable markets: compared with their religious peers, women living in a market with fewer co-religionists had lower odds of marriage. I tested alternative specifications of religious concentration, including adding a squared term and using a categorical version of the variable. The linear term fit the model best. Thus, for women, the results supported Hypothesis 1 and 6: religious concentration was positively associated with marriage timing but only among women in their mid to late twenties.

Most other market variables, including race-ethnic composition, were not associated with marriage for women. The proportion of unmarried men employed full-time, which was negatively associated with young women’s entry into first marriage. This could reflect that many men at these early ages—especially those pursuing college degrees and high-paying careers—are delaying marriage until later in the life course. It could also be capturing aspects of women’s economic prospects. Research has found that women’s employment opportunities are negatively associated with marriage rates in the aggregate (Ellwood & Jencks 2004).

Table 4 presents marriage timing results for men. With the exception of large differences between non-Hispanic whites and blacks at early ages, there were not significant differences in first marriage between men of different race-ethnic groups. Educational attainment and earnings were positively associated with first marriage for men, but, unlike for women, they were significant at both life course stages. Weekly service attendance was significantly associated with men’s marriage but there were few differences between religious groups, net of controls. Living in the Northeast was negatively associated with marriage for young men but not older men. For men in both age groups, having a child or a partner who was pregnant increased the chances of marriage.

Table 4.

Coefficients from event history logistic regression models of transition into first marriage for men by life course stage (ages 18–23 and 24–31), weighted

	Ages 18–23			Ages 24–31
	B	SE B	OR	B	SE B	OR
Religious market concentration	−0.06	0.05	0.95	0.10*	0.05	1.10
Religious affiliation (ref: Evangelicals)
Mainline Protestant	0.12	0.13	1.13	0.17	0.12	1.19
Catholic	−0.39**	0.12	0.68	0.06	0.12	1.07
Jewish	−0.13	0.51	0.88	−0.38	0.38	0.68
Mormon	0.18	0.32	1.20	0.34	0.45	1.40
Unaffiliated	−0.22	0.15	0.80	−0.08	0.14	0.92
Other	−0.42	0.36	0.66	−0.18	0.34	0.84
Attends services weekly	0.74***	0.12	2.10	0.76***	0.10	2.15
Parental religiosity index	0.02	0.03	1.02	−0.01	0.03	0.99
Race-ethnicity (ref: NH White)
NH Black	−1.35***	0.26	0.26	0.45	0.81	1.56
Hispanic	0.04	0.18	1.04	−0.23	0.24	0.80
Other	−0.11	0.27	0.90	−0.51	0.53	0.60
Educational attainment (ref: high sch.)
Less than high school	−0.29*	0.14	0.75	−0.28	0.18	0.76
Associate’s degree	0.40†	0.24	1.49	0.39*	0.16	1.48
Bachelor’s degree or more	0.53†	0.27	1.70	0.46***	0.11	1.58
Enrolled in school (=1)	−0.11	0.11	0.89	0.25†	0.13	1.28
Employment status (ref: not employed)
Part-time	0.20	0.15	1.22	0.24	0.18	1.27
Full-time	0.90***	0.15	2.46	0.77***	0.15	2.16
Annual earnings	0.06***	0.02	1.06	0.04*	0.02	1.04
Other market characteristics
Sex composition (% opposite-sex)	0.01	0.02	1.01	−0.03	0.02	0.97
Educational concentration	−0.00	0.01	1.00	−0.00	0.01	1.00
Economic attractiveness	−0.00	0.01	1.00	0.00	0.01	1.00
Religious diversity index	−0.01	0.09	0.99	−0.19†	0.10	0.83
Cultural conservatism (% vote for Bush 2004)	0.01	0.01	1.01	−0.01†	0.01	0.99
Constant	−3.59	2.32		0.48	2.41
Person-years	23928			16099

Note: All models control for age, year, parent’s educational attainment, family structure at age 12, first birth status, metropolitan and regional residence, local age structure, race-ethnic composition, educational composition, annual unemployment rate, % urban, and total population size (logged). Standard errors are adjusted for clustering of respondents within commuting zones using Stata’s <cluster> command.

^†p<0.1

^*p<0.05

^**p<0.01

^***p<0.001

Similar to the results for women, religious concentration was significantly associated with first marriage for older but not younger men. Relative to a respondent’s religious peers, living in a marriage market with one standard deviation greater concentration of co-religionists was associated with an 11 % increase in the annual risk of marriage for men ages 24–31 (exp[0.100] = 1.105). This supports Hypotheses 1 and 6 among men. In pooled models (which included both genders), the interaction term between religious concentration and gender was not significant. This does not support Hypothesis 3, which expected women’s risk of marriage to be more sensitive to the supply of religiously similar partners than men’s risk of marriage.

Hypotheses 4 and 5 expected that young adults from more conservative religious traditions and those with greater religious commitment would show a stronger association between religious concentration and marriage because they value partner’s religion more highly. Results of marriage timing models for women and men (by age), shown in Table A2 of the online appendix, were mixed. There was no statistically significant interaction between religious affiliation and religious concentration for younger women and men or for older young-adult women. There were significant differences by religious traditions for older men. The association between religious concentration and marriage was stronger for Catholics, Mormons, and Unaffiliated men compared to other groups. A one standard deviation increase in the concentration of co-religionists was associated with a 21% increase in the annual risk of marriage for Catholics, a 33% increase for unaffiliated men, and a 58% increase for Mormons. I also tested the interaction between religious concentration and weekly attendance but did not find support for Hypothesis 5.

Assortative Mating

Table 5 shows results from multinomial logistic regression event history models predicting partner’s religion for women and men ages 24–31. The first column in each set contrasts marriage to a spouse of the same religion with marriage to a spouse of a different religion (homogamy vs. heterogamy, or assortative mating conditional on marriage), while the second column contrasts religious homogamy to remaining unmarried (the third possible contrast, not shown, compares religious heterogamy to remaining unmarried). The first column is similar to previous studies of religious assortative mating (Lehrer 1988; Sherkat 2004).

Table 5.

Coefficients from event history multinomial logistic regression models of spouse’s religion by gender, ages 24–31, weighted

	Women, ages 24–31						Men, ages 24–31
	Religious homogamy vs. heterogamy			Religious homogamy vs. unmarried			Religious homogamy vs. heterogamy			Religious homogamy vs. unmarried
	B	SE B	RRR	B	SE B	RRR	B	SE B	OR	B	SE B	RRR
Religious market concentration	0.55***	0.10	1.74	0.42***	0.08	1.52	0.57***	0.11	1.78	0.37***	0.09	1.45
Religious affiliation (ref: Evangelicals)
Mainline Protestant	−1.06***	0.26	0.35	−0.46*	0.19	0.63	−1.63***	0.26	0.20	−0.80***	0.23	0.45
Catholic	−0.43†	0.25	0.65	−0.36†	0.19	0.70	−0.32	0.25	0.73	−0.14	0.17	0.87
Jewish	−0.26	0.78	0.77	−0.46	0.51	0.63	0.07	0.67	1.07	−0.57	0.42	0.56
Mormon	−1.41*	0.64	0.24	−0.87†	0.47	0.42	−0.63	0.79	0.53	−0.22	0.52	0.80
Unaffiliated	−1.20**	0.39	0.30	−0.88**	0.29	0.42	−1.13***	0.31	0.32	−0.64*	0.25	0.53
Other	−1.62†	0.92	0.20	−1.71*	0.77	0.18	−0.81	0.57	0.45	−0.51	0.51	0.60
Attends services weekly	0.79**	0.25	2.20	0.80***	0.14	2.22	0.29	0.20	1.34	0.90***	0.13	2.45
Parental religiosity index	−0.03	0.06	0.97	−0.02	0.04	0.98	0.10†	0.06	1.11	0.06	0.05	1.07
Educational attainment (ref: high sch.)
Less than high school	0.23	0.39	1.26	0.05	0.25	1.06	−0.83*	0.37	0.44	−0.82**	0.30	0.44
Associate’s degree	0.23	0.31	1.26	0.50*	0.23	1.64	−0.03	0.33	0.97	0.48*	0.23	1.62
Bachelor’s degree or more	0.16	0.22	1.18	0.61***	0.17	1.84	−0.03	0.21	0.97	0.55**	0.17	1.73
Enrolled in school (=1)	0.46†	0.24	1.59	0.20	0.19	1.22	0.15	0.25	1.16	0.30†	0.18	1.34
Employment status (ref: not employed)
Part-time	0.35	0.26	1.42	0.10	0.18	1.10	0.67†	0.36	1.96	0.55*	0.27	1.74
Full-time	0.35	0.25	1.42	0.17	0.19	1.19	0.57	0.37	1.76	1.01***	0.25	2.74
Annual earnings	−0.03	0.03	0.97	0.03	0.02	1.04	−0.07†	0.04	0.94	0.02	0.02	1.02
Other market characteristics
Sex composition (% opposite-sex)	0.04	0.05	1.04	−0.00	0.04	1.00	−0.12**	0.04	0.88	−0.09**	0.03	0.92
Educational concentration	0.03	0.02	1.03	0.00	0.01	1.00	−0.00	0.02	1.00	−0.00	0.01	1.00
Economic attractiveness	−0.00	0.02	1.00	0.00	0.01	1.00	0.01	0.02	1.01	0.01	0.01	1.01
Religious diversity index	−0.35†	0.19	0.70	−0.26†	0.14	0.77	−0.02	0.20	0.98	−0.20	0.17	0.82
Cultural conservatism (% vote for Bush 2004)	−0.01	0.01	0.99	0.00	0.01	1.00	0.02	0.02	1.02	−0.01	0.01	0.99
Constant	4.50	5.76		0.50	3.94		7.56	5.17		3.90	3.74
Person-years	13504						15751

Note: All models control for race-ethnicity, age, year, parent’s educational attainment, family structure at age 12, first birth status, metropolitan and regional residence, local age structure, race-ethnic composition, educational composition, annual unemployment rate, % urban, and total population size (logged). Standard errors are adjusted for clustering of respondents within commuting zones using Stata’s <cluster> command. Race-ethnic differences in assortative mating were not significant; not shown to conserve space.

^†p<0.1

^*p<0.05

^**p<0.01

^***p<0.001

In the first column, Evangelical women and men were more likely to marry assortatively on religion than their Mainline, Catholic, and Unaffiliated peers. Parental religiosity was positively associated with religious homogamy for men. Supporting Hypothesis 2, religious market concentration was positively associated with assortative mating for women. Women in markets with more co-religionists were more likely to marry a spouse who shared their religion (assortative mating). There was a similarly strong relationship between religious market concentration and religious homogamy for men. In pooled models, the difference in the association between religious concentration and assortative mating among women was not statistically different from its association among men. In contrast to the marriage timing models, religious market concentration was also associated with assortative mating for younger men and women. This highlights the utility of search theory’s approach: conditional on finding a partner, younger men and women, ages 18–23, were more likely to find a religious match in market with higher concentrations of religious similar partners, but they were not more likely to enter marriage. Together, this suggests religious similarity is a more salient partner trait among older young adults compared to younger adults.

First Union: Marriage vs. Cohabitation

I also estimated models of first-union formation for men and women, contrasting marriage to cohabitation. The results for young adults ages 24–31 presented in Table 6 provided mixed support for Hypothesis 7, which anticipated that religious concentration would be more strongly associated with entry into marriage compared to cohabitation. For women, religious concentration was positively associated with marriage over cohabitation, but the association was only marginally significant (p = 0.076). Religious concentration was not associated with remaining un-partnered over cohabitation. For men, religious concentration was not associated with either marriage over cohabitation or remaining un-partnered over cohabitation.

Table 6.

Coefficients from event history multinomial logistic regression models of transition to first union by union type for men and women, ages 24–31, weighted

	Women, ages 24–31						Men, ages 24–31
	Marriage vs. cohabitation			Unmarried vs. cohabitation			Marriage vs. cohabitation			Unmarried vs. cohabitation
	B	SE B	RRR	B	SE B	RRR	B	SE B	RRR	B	SE B	RRR
Religious market concentration	0.23†	0.12	1.10	0.09	0.06	1.26	0.02	0.11	0.97	−0.04	0.05	1.02
Religious affiliation (ref: Evangelicals)
Mainline Protestant	0.19	0.28	1.04	0.04	0.16	1.21	−0.31	0.29	0.95	−0.05	0.15	0.73
Catholic	−0.71**	0.27	0.95	−0.05	0.16	0.49	−0.39	0.29	0.84	−0.18	0.14	0.68
Jewish	0.49	0.65	0.87	−0.13	0.33	1.62	−0.22	0.61	1.35	0.30	0.38	0.80
Mormon	1.60*	0.75	3.42	1.23†	0.65	4.97	0.46	0.67	1.05	0.05	0.51	1.59
Unaffiliated	−0.77†	0.40	0.77	−0.27	0.16	0.46	−0.79*	0.37	0.92	−0.08	0.15	0.46
Other	0.85	0.90	2.23	0.80†	0.42	2.33	0.10	0.62	1.37	0.31	0.31	1.10
Attends services weekly	1.79***	0.25	2.32	0.84***	0.19	5.97	2.00***	0.25	1.74	0.56**	0.17	7.40
Parental religiosity index	0.06	0.09	0.98	−0.02	0.04	1.06	−0.03	0.06	1.00	0.00	0.03	0.97
Educational attainment (ref: high sch.)
Less than high school	0.32	0.71	1.16	0.14	0.25	1.38	−0.65*	0.31	1.16	0.15	0.17	0.52
Associate’s degree	0.58	0.37	1.14	0.13	0.21	1.79	0.23	0.31	0.96	−0.04	0.18	1.26
Bachelor’s degree or more	0.26	0.28	0.95	−0.05	0.13	1.30	0.25	0.19	0.80	−0.22*	0.10	1.29
Enrolled in school (=1)	0.19	0.28	1.16	0.15	0.13	1.21	0.80***	0.21	1.05	0.05	0.13	2.23
Employment status (ref: not employed)
Part-time	−0.04	0.32	0.77	−0.26	0.16	0.96	−0.31	0.32	0.66	−0.42*	0.17	0.74
Full-time	0.40	0.34	0.81	−0.21	0.16	1.49	0.44	0.29	0.55	−0.61***	0.15	1.55
Annual earnings	0.03	0.04	1.00	0.00	0.02	1.03	−0.02	0.03	0.99	−0.01	0.02	0.98
Other marriage market characteristics
Sex composition (% opposite-sex)	−0.05	0.06	0.99	−0.01	0.03	0.95	−0.06	0.04	0.97	−0.03	0.02	0.95
Educational concentration	0.01	0.02	1.01	0.01	0.01	1.01	−0.02	0.03	1.00	0.00	0.01	0.98
Economic attractiveness	−0.03†	0.02	1.00	0.00	0.01	0.97	−0.00	0.02	1.00	−0.00	0.01	1.00
Religious diversity index	−0.22	0.20	0.97	−0.03	0.11	0.81	−0.24	0.19	0.94	−0.06	0.09	0.79
Cultural conservatism (% vote for Bush 2004)	0.01	0.01	1.01	0.01	0.01	1.01	0.01	0.01	1.01	0.01†	0.01	1.01
Constant	−5.31	5.25		−0.84	2.37		4.52	4.62		3.73†	2.16
Person-years	6444						6444

Note: All models control for race-ethnicity, age, year, parent’s educational attainment, family structure at age 12, first birth status, metropolitan and regional residence, local age structure, race-ethnic composition, educational composition, annual unemployment rate, % urban, and total population size (logged). Standard errors are adjusted for clustering of respondents within commuting zones using Stata’s <cluster> command. Race-ethnic differences in union type (marriage v. cohabitation) were not significant; not shown to conserve space.

^†p<0.1

^*p<0.05

^**p<0.01

^***p<0.001

Robustness Checks

As described in the Methods section, it was not possible to calculate race-specific measures of local religious composition using the RCMS data. RCMS data also had poor coverage of Black Protestants congregations. I addressed these shortcomings by (1) estimating all models separately for non-Hispanic whites and non-Whites, and (2) creating a synthetic measure of the race-specific religious concentration using state-level data.

The results are summarized in Table 7. Each panel shows coefficients for religious concentration from models estimated separately for non-Hispanic whites and non-whites by gender (ages 24–31; results for ages 18–23 were not significant). The first model includes the non-race-specific measure of religious concentration used in previous models; the second includes the race-specific measure. The first panel shows results for marriage timing, the second for assortative mating (only homogamy vs. heterogamy contrast), and the third for first union type. The results for women show that patterns observed in the main models were driven largely by whites. There was a strong positive association between religious concentration and marriage timing, assortative mating, and first union type for white women. In contrast, there was no association between religious concentration and marriage timing or first union type for non-whites using either measure. Coefficients in assortative mating models were at or near statistical significance, but effect size was more muted for non-whites compared to whites. In pooled models, the interaction between religious concentration and race-ethnicity showed a stronger effect for whites compared to other groups in marriage timing and first union models.

Table 7.

Coefficients for religious market concentration from models by race-ethnicity (non-Hispanic whites and other groups) for women and men ages 24–31, using original and race-specific measures of religious concentration, weighted

	Women, ages 24–31			Men, ages 24–31
	B	SE B	OR/ RRR	B	SE B	OR/ RRR
Marriage timing
Non-Hispanic Whites only
M1: Religious concentration	0.12*	0.05	1.13	0.13*	0.06	1.13
M2: Race-specific religious concentration	0.12*	0.05	1.12	0.08	0.06	1.08
Non-Whites only (Blacks, Hispanics, and others)
M1: Religious concentration	0.02	0.08	1.07	0.05	0.08	1.05
M2: Race-specific religious concentration	−0.01	0.09	0.99	−0.01	0.06	0.99
Assortative mating (homogamy vs. heterogamy)
Non-Hispanic Whites only
M1: Religious concentration	0.55***	0.13	1.74	0.53***	0.13	1.70
M2: Race-specific religious concentration	0.53***	0.13	1.70	0.50***	0.13	1.64
Non-Whites only (Blacks, Hispanics, and others)
M1: Religious concentration	0.35†	0.19	1.41	0.57***	0.16	1.76
M2: Race-specific religious concentration	0.39*	0.18	1.48	0.38**	0.15	1.46
First union (marriage vs. cohabitation)
Non-Hispanic Whites only
M1: Religious concentration	0.35*	0.16	1.42	0.02	0.15	1.02
M2: Race-specific religious concentration	0.31*	0.16	1.37	−0.00	0.14	1.00
Non-Whites only (Blacks, Hispanics, and others)
M1: Religious concentration	−0.00	0.17	1.00	−0.03	0.16	0.97
M2: Race-specific religious concentration	−0.02	0.21	0.98	−0.09	0.12	0.91

Note: All models control for same set of covariates in main models. Standard errors are adjusted for clustering of respondents within commuting zones.

^†p<0.1

^*p<0.05

^**p<0.01

^***p<0.001

Results for men by race-ethnicity were more mixed. Results for marriage timing using the unadjusted measure of religious concentration were stronger for non-Hispanic white men than non-whites. However, the results for white men using the race-specific measure were more muted and not statistically significant for marriage timing. However, the effect of religious concentration was stronger and statistically significant for Catholic, Mormon, unaffiliated, and Jewish non-Hispanic white men. Results from assortative mating models were similar across race-ethnic groups and both measures of religious concentration. Religious concentration was not significantly associated with first union type for white or non-white men. Overall, these analyses provide reassurance that the results are not heavily influenced by the shortcomings of the RCMS data, at least for non-Hispanic whites. In the discussion section, I propose possible reasons for race-ethnic differences in the marriage timing models.

Finally, while it was beyond the scope of this study to compare the importance of religion relative to other partner traits, the association between religious concentration and religious homogamy (vs. heterogamy) among married respondents held after controlling for race-ethnic and educational homogamy. This suggests that religious concentration was not a proxy for the availability of racially or educationally similar partners. See online supplement for more detail.

Discussion

This study used search theory to clarify the importance of religious assortative mating for marriage decisions among contemporary young adults in the United States. Overall, I found support for search theory’s main hypotheses (Hyp. 1 and 2) but the results varied by life course stage. Among 24–31 year olds, young adults in favorable markets for religious assortative mating were more likely to enter marriage and to marry a religiously similar partner relative to their religious peers. Likewise, women and men in less favorable markets were more likely to delay marriage, on average, rather than marry outside their religion. This pattern did not vary by gender (no support for H3). Results for older young adults support the idea that religious similarity is important in the marriage market, affecting marriage timing and partner choice.

Consistent with Hypothesis 6, there was no association between religious market concentration and marriage timing among younger adults ages 18–23. On the one hand, this could indicate that religious homogamy is not a driving factor behind marriage decisions in early adulthood. At these young ages, religious participation tends to be low and most young adults take a more casual approach to romantic relationships. On the other hand, the lack of association between market concentration and marriage timing in early adulthood could also reflect the changing boundaries of the marriage market over the life course. Young adults looking to marry at these ages may be more likely to find partners in their local neighborhood rather than the wider county or commuting zone of residence (Warner et al. 2011; Laumann et al. 2004). In addition, the social composition of settings like churches or schools may capture marriage-market conditions better than geographic boundaries (Kalmijn 1998; McClendon et al. 2014). Unfortunately, measuring the religious composition of more narrowly tailored marriage markets is not possible with available data.

The strong association between the religious composition of commuting zones and marriage behavior for older young adults suggests not only that religion is important to prospective partners at these ages, but also that the boundaries of the marriage market widen in the years following school completion. As more young adults begin actively searching for a spouse, they appear to cast a wider geographic net, making the social composition of the larger market more consequential for marriage chances. Ironically, online dating may be helping many young adults in their search for a religiously similar partner. While online dating allows people to circumvent traditional gatekeepers and meet a more diverse set of romantic partners, online dating also makes it easier to act on and realize one’s partner preferences (Rosenfeld & Thomas 2012). For instance, studies of messaging behavior on popular dating sites find that race-ethnic boundaries and hierarchies in the dating market persist in cyberspace (Lundquist & Lin 2013). Online dating sites geared toward specific religious groups may make it even easier for young adults who are seeking a religiously similar spouse to find one in their local area.

The association between market concentration and marriage timing did not vary by religious commitment or by religious tradition for women. The association did vary by religious tradition for men (stronger for Mormons, Catholics, and the unaffiliated), but the pattern was not wholly consistent with Hypothesis 4, which expected a stronger association among men from conservative religious traditions. Together, these findings suggest that religion’s appeal on the marriage market may have more to do with cultural compatibility between partners than concerns about doctrinal differences or religious pressures to marry within the faith. Even as fewer Americans are identifying with a religion and attitudes toward intermarriage grow more accepting, religion continues to shape people’s beliefs and values about family life and to act as an important source of identity. While religious similarity may not register as a top concern for young adults in the abstract (South 1991; Raley & Bratter 2004; Smith & Denton 2005), it nevertheless appears to be an attractive trait in the actual marriage market, at least at these ages. It is likely that young adults gravitate towards partners who share their values, cultural identity, and politics – all of which may be captured by a partner’s religious or non-religious background.

This cultural matching perspective on assortative mating contrasts with other perspectives that emphasize competition and mercenary mate selection on the marriage market. Prior applications of search theory have focused on socioeconomic characteristics of partners, like income, employment, and education (Lichter et al. 1992; Lewis & Oppenheimer 2000; Schwartz 2010; Sweeney & Cancian 2004). While scholars disagree over whether these forms of assortative mating reflect sorting processes that are characterized by competition or matching (Schwartz 2013), this study calls attention to how religious compatibility and cultural matching, more broadly, are important considerations for many young adults when choosing partners. A promising area for future research would be to examine how matching on the basis of cultural compatibility intersects with competition for economically attractive mates on the market.

There was also significant racial-ethnic variation. Among 24–31 year olds, religious concentration was positively associated with marriage timing for non-Hispanic whites but not for blacks, Hispanics, and other groups. One reason for race-ethnic variation could be error in the measurement of religious market composition for non-whites, especially for blacks. However, there are other potential explanations. Whites tend to have more education and better economic prospects than other groups, which improve their marriage chances (Raley, Sweeney, & Wondra 2015). In this context, it is possible that whites can afford to be choosy and consider religion when searching for a mate because they are more attractive to potential partners and they have a larger supply of economically viable, racially similar mates from which to choose. For others, finding an economically attractive partner of the same race is difficult, much less a religiously similar one.

This study has a few other limitations. First, because the NLSY97 cohort is still young, this analysis describes relatively early marriage for today’s young adults. The findings may not be generalizable to marriage-market dynamics later in the life course. Second, measurement of partner’s religion was retrospective for many respondents, which may have led to underestimates of assortative mating if large numbers of spouses switched religions to match their partner after marriage. Third, it was beyond the scope of this study to compare the relative importance of religion to other partner characteristics in the assortative mating process. Given the results of this analysis, which indicate that religion is a salient partner trait on the marriage market for many young adults, future research should investigate this question more closely.

Findings from this study offer a different perspective on religious assortative mating in the United States. Much has been written about the rise of interfaith marriage since the 1950s (Putnam and Campbell 2010). There is evidence of this decline among NLSY97 respondents: half of young adults who entered marriage prior to age 31 married a spouse who did not share their religion. Yet, trends in religious homogamy among prevailing marriages only tell us so much about social boundaries that divide the marriage market. By looking at the connections between local opportunities for assortative mating and entry into marriage, this study finds that partner’s religion continues to matter to young adults, at least for non-Hispanic whites, as they search for potential spouses. Furthermore, the strong association between the supply of religiously similar partners and assortative mating highlights that current interreligious marriage patterns reflect not just changes in preferences but also spatial variation in access to partners.