Eight Decades of Educational Assortative Mating: A Research Note

Introduction

The degree of educational assortative mating in societies is a key indicator of the social distance between socioeconomic groups and is a mechanism for the unequal distribution of resources across households and children. Educational assortative mating is thus one link in the chain that transmits parental characteristics to children across generations. Schwartz and Mare (2005, hereafter “SM”) showed a long-run increase in the educational resemblance of spouses from 1960 to 2003. They found that both the most and least educated Americans had become less likely to intermarry across educational boundaries. These findings sparked new interest in the connection between assortative mating and inequality (e.g., Boertien and Permanyer 2019; Breen and Salazar 2011; Shen 2021). In this research note, we update SM’s findings with data through 2020. There has been considerable social and economic change in the United States since 2003, with many potential implications for patterns of educational intermarriage. We ask whether the last two decades have seen a continued rise in educational homogamy—thus examining eight decades of educational assortative mating in the U.S.—and expand on SM’s study by examining heterogeneity by race, ethnicity, nativity, and between same- and different-sex couples.

There is reason to expect educational homogamy has continued to increase. Economic inequality continued to rise through the early 2010s and this is closely linked to increased returns to educational attainment (Aeppli and Wilmers 2022; Autor 2014). When the returns to education are high, past theory and research predict lower rates of intermarriage given that the costs of “marrying down” are high (Fernández, Guner, and Knowles 2005; Torche 2010). Education may therefore increasingly shape the incentives and opportunities for selecting potential partners. Income segregation across neighborhoods increased between the 1970s and mid-2000s, although may have declined since then (Leung-Gagné and Reardon 2023; Reardon et al. 2018) and income segregation in the workplace increased between the early 1970s and 2013 (Song et al. 2019). Spatial and workplace segregation may further reduce opportunities for matching among individuals with different levels of education.

Other trends point to stagnation or even a potential decline in educational homogamy. First, increases in the age at marriage may mean a lengthening of the time between school completion and marriage, reducing the influence of education on partner choice (Mare 1991; Schwartz 2013). Second, educational homogamy may be declining as higher education becomes less elite and more diverse. Nearly all adolescents now aspire to complete college, and many more successfully attend than in the past (Goyette 2008; Radford et al. 2018). Adults also have increasingly returned to college later in life (Denice 2017), which may have resulted in more individuals with college degrees but whose social ties lead them to form partnerships with less-educated individuals compared to the college-educated of earlier decades, thus increasing educational heterogamy (Armstrong and Hamilton 2021; King 2021). Third, the continued rise in women’s education relative to men’s means that there is increasing demographic pressure for women to form unions with men who have less education than themselves (Van Bavel, Schwartz, and Esteve 2018). This trend may reduce educational homogamy and increase hypogamy mechanically, that is, without any changes in matching preferences, but people may also adjust their expectations due to marriage market constraints (Corti and Scherer 2021; Esteve et al. 2016; Fong 2023; Grow and Van Bavel 2015; Han 2022).

Other factors have more ambiguous potential influences on patterns of educational homogamy. For example, increases in gender egalitarianism may increase or decrease educational homogamy. Highly educated men may be more likely to seek highly educated women today than in the past, thus increasing educational homogamy, but highly educated women may be less concerned about the earnings prospects of male partners, thus decreasing educational homogamy (Fernández et al. 2005; Press 2004). Meeting patterns have also shifted dramatically, with uncertain effects. Couples increasingly meet online rather than through traditional intermediaries like family, friends, schools, or the workplace (Rosenfeld and Thomas 2012). Internet dating may allow individuals to better match on observable characteristics like educational attainment but the traditional intermediaries it displaced were also strong engines of educational homogamy. Consistent with the ambiguous predictions, some studies have found little evidence of differences in assortative mating between those who met online versus not online (Potarca 2020, 2021; Kornrich, Sabino, & Robbins 2024), but other studies find lower levels of homogamy among those who met online (Potarca 2017; Thomas 2020).

Trends in educational assortative mating also likely vary by race, ethnicity, nativity, and for same- versus different-sex couples. We expand on SM to consider this heterogeneity. Racial differences in union formation at all levels of education have grown in recent decades (Raley, Sweeney, & Wondra, 2015), and this may also be the case for educational homogamy. Variation in gender gaps in education by race, ethnicity, and nativity may also produce differences in educational assortative mating (Coley 2001; McDaniel et al. 2011). The United States has become increasingly multi-racial and multi-ethnic as the post-1965 wave of immigrants expanded. The rising population of immigrants may have altered overall trends in educational homogamy if immigrants’ assortative mating tendencies differ from the native-born. Same-sex couples tend to be less homogamous on a range of social and demographic characteristics, owing either to different opportunities for finding partners, or to differing partner preferences (see Reczek 2020 for a recent review). Broad social change such as the national legalization of same-sex marriage in 2015 may have shaped partnering patterns among same-sex couples. We describe recent trends in educational assortative mating among same-sex couples in this context.

Our results have implications for understanding trends in the social distance between education groups, gendered patterns of relationship formation, and the potential role of assortative mating in exacerbating economic inequality. SM found declining intermarriage between educational groups between 1960 and 2003, consistent with growing social and economic distance by education and the erosion of male status-dominance norms. Given these patterns, they speculated that educational homogamy could potentially be an engine of inequality as economically advantaged men and women increasingly married one another. Although subsequent research has largely failed to find effects of increased educational homogamy on economic inequality (e.g., Boertien and Permanyer 2019; Breen and Salazar 2011; Schwartz, Wang, and Mare 2021), their findings raised the possibility that various forms of inequality within and between generations may be increasing in part due to intensifying marital sorting.

We update the empirical trend through 2020, thereby providing up-to-date context for new scholarship investigating the links between marital sorting and other forms of inequality. By contrast to SM, we find very different recent patterns of educational assortative mating. The increase in educational homogamy noted by SM stalled around 1990 and began reversing in the 2000s. We find that a key factor explaining this new pattern is women’s increasing tendency to marry men with less education than themselves. We describe how our findings shift the conversation about the potential implications of trends in assortative mating in the United States and other countries.

Data and Methods

We construct a dataset of representative cross-sections of different-sex married couples from the U.S. decennial censuses between 1940 and 2000 and the American Community Survey (ACS) from 2001 to 2020 (Ruggles et al. 2022). For same-sex couples, we rely on ACS data from 2008 to 2020, years for which issues of gender misreporting are substantially reduced and the identification of same-sex couples is more accurate (Ruggles et al. No date; U.S. Census Bureau 2013). To further reduce the possibility of erroneously classifying different-sex couples as same-sex couples, we follow past research and drop (1) same-sex couples in which the sex of either partner was “allocated,” that is, changed from its original value and (2) same-sex couples who reported marrying prior to 2004, when Massachusetts became the first state in which same-sex couples legally married (Gates 2015). We include both married and cohabiting same-sex couples given the change in the law governing same-sex marriage during this period and also include both cohabiting and married different-sex couples in our comparisons with same-sex couples.¹ We restrict our samples to couples in which women (among different-sex couples) or householders (among same-sex couples) are age 18–40.

We cross-classify couples by educational attainment in six categories that are inclusive of the changing distribution of education over time: fewer than 10 years of schooling, 10–11 years, 12 years (high school graduate), 13–15 years (some college or associate’s degree), 16 years (bachelor’s degree), and 17 or more years (graduate or professional degree). SM did not separate bachelor’s from graduate degrees and, given the growing proportion of the population with these degrees, this is an additional contribution of our article. Although we examine prevailing marriages in most models, in some, we restrict the sample to newlyweds (marriages beginning within two years prior to the survey). Throughout, we measure spouses’ education at the time of the survey. It is possible that respondents received more education later in life compared to when they married. Our results for newlyweds, however, are largely free of any such effects because they rely on survey measurements timed shortly after marriage.

We examine trends in educational assortative mating using log-linear models following SM. Dramatic changes in the distribution of spouses’ education since 1940 can affect measures of the association simply due to changes in the size of education categories. Our primary objective is to study trends in spouses’ education associations independent of changes in spouses’ education distributions. Log-linear models are therefore appropriate given that they control for changes in the marginal distributions of spouses’ education (Powers and Xie 2008:97–99). Like SM, our baseline model takes the form

$$\text{log}({\mu }_{ijk}/t_{ijk})=\lambda +{\lambda }_i^H+{\lambda }_j^W+{\lambda }_k^Y+{\lambda }_{ij}^{HW}+{\lambda }_{ik}^{WY}+{\lambda }_{jk}^{HY},$$ (1)

where H is husband’s education (i = 1, …, 6), W is wife’s education (j = 1, …, 6), and Y is year (k = 1940, …, 2020). On the left-hand side, μ_ijk is the expected number of marriages for each combination of husband’s education i, wife’s education j, and year k.² The models are the same for same-sex couples except that they are classified by householders’ and partners’ characteristics. We focus on the results of log-linear models in this article but show descriptive statistics in Online Appendix (trends in the percentage of marriages that are homogamous and in which wives have more education than their husbands [Appendix Table S1]; trends in the joint distribution of husbands’ and wives’ education [Appendix Table S2]; and trends in the gender difference in years of schooling among men and women aged 25–40 [Appendix Figure S1]).³

Following SM, we add complexity to equation (1) to investigate time trends in assortative mating. We first examine changes in the odds of homogamy by adding a homogamy parameter (which equals 1 where H=W and 0 otherwise) that varies by year. We next estimate a crossings model, which parameterizes the odds of crossing each adjacent educational boundary.⁴

We also include hypogamy parameters for different-sex couples—identifying couples in which wives have a higher education category than husbands—in addition to the homogamy parameters. The inclusion of hypogamy parameters changes the interpretation of the exponentiated homogamy coefficients, such that they describe the change in the odds of homogamy relative to hypergamy (husbands have more education than their wives), rather than relative to heterogamy (spouses have different education categories). Because we present trends in odds, the y-axes of all our figures are shown in the log scale.

Finally, to examine heterogeneity in assortative mating, we include interactions by wife’s race/ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, non-Hispanic Asian/Pacific Islander), nativity (foreign or native born), and an indicator for same-sex male, same-sex female, and different-sex couples in our models.⁵

To address the sensitivity of our results to alternative specifications, the Online Appendix shows fit statistics for different models of trends (Appendix Table S3), coefficients for alternative single-parameter trends (Appendix Figure S2), trends using different classifications of husbands’ and wives’ education (Appendix Figure S3), and trends with allocated education values dropped (Appendix Figure S4). Our broad conclusions hold across these specifications.

Results

Educational homogamy among different-sex married couples

Contrary to the expectation that educational homogamy would continue to increase Figure 1 shows that the odds of educational homogamy among U.S. different-sex married couples held steady at around 4:1 beginning around 1990 and declined somewhat since the early 2000s.⁶ We find further evidence that homogamy has declined from its peak when we restrict our sample to newlyweds: the odds of homogamy among the newly married fell from 3.7:1 in 1980 to 3.3:1 in 2020.

Figure 1:

Odds of Homogamy, Different-Sex Prevailing Marriages and Newlyweds, 1940–2020

Notes: Wives aged 18–40. Education categories are <10, 10–11, 12, 13–15, 16, 16+.

Sources: 1940–2000 U.S. decennial census data and 2001–2020 American Community Survey data (IPUMS).

To further explore the underlying patterns of this stability and decline in educational homogamy, we estimate crossings models, which show the odds of marriages across adjacent educational boundaries (Figure 2). Like SM, we observe a long-term decline in intermarriage between college graduates and those with some-college as well as among those at the bottom of the education distribution, consistent with the pulling away of the economic and social elite from the rest of the U.S. population and the falling behind of those with very little education (Keister, 2014; McLanahan, 2004). Unlike SM, we show that the odds of intermarriage between college graduates and those with advanced degrees is quite high, indicating a weak barrier, although these odds declined as well between 1980 and roughly 2000. Consistent with others’ findings of the importance of a college degree for family and economic patterns (e.g., Lundberg, Pollak, and Stearns 2016), our results show that this is also true for educational assortative mating. A college degree is the dividing line across which it is most difficult to intermarry.

Figure 2:

Odds of Crossing Educational Boundaries, Different-Sex Prevailing Marriages, 1940–2020

Notes: Wives aged 18–40. Odds of crossing adjacent educational boundaries relative to homogamy. Estimates use 3-year intervals between 2000–2020 for legibility. The equivalent results using 1-year intervals are available in Online Appendix Figure S5.

Sources: 1940–2000 U.S. decennial census data and 2001–2020 American Community Survey data (IPUMS).

Updating trends beyond SM, intermarriage between 2000 to 2010 was relatively stable with the exception of a continuation of the decades-long weakening of the barrier to intermarriage between high school graduates and those with some college. But beginning around the 2010s, a new pattern emerged, one of increasing intermarriage across all educational boundaries. This is a notable and perhaps surprising finding.

The increasing tendency for women to marry less educated men (hypogamy) has occurred around the world (Esteve et al., 2016) and may be an important component of the stagnation and decline in homogamy in recent decades. Figure 3 shows evidence from our log-linear models supporting this. Both the odds of hypogamy (Panel A) and homogamy (Panel B) relative to hypergamy increased between 1970 and about 2010. Thus, if the odds of hypogamy had not increased, the odds of homogamy would have continued to increase until at least the 2010s rather than stabilizing around 1990 as observed. The decline in the odds of homogamy since the 2010s is due to both a continued increase hypogamy (Panel C) and a slight resurgence of hypergamy (Panel B) at the expense of homogamy.⁷ Note that Panels A to C are presented on different y-axis scales for legibility but we also present the three trends on the same scale in Panel D. Although the levels are much different, the rise of hypogamy at the expense of homogamy and hypergamy is still evident. It is interesting to note that the decline of homogamy relative to hypogamy and hypergamy since the 2010s is consistent with the timing of reduced socioeconomic inequality in the United States since 2012 (Aeppli and Wilmers 2022). Nevertheless, the main finding is that the odds of homogamy stopped increasing in 1990 and have declined primarily because of the increasing odds that wives have more education than their husbands.⁸

Figure 3:

Odds of Homogamy (W=H), Hypogamy (W>H), and Hypergamy (W<H), Different-Sex Prevailing Marriages, 1940–2020

Notes: Wives aged 18–40. W=Wives’ education. H=Husbands’ education. Education categories are <10, 10–11, 12, 13–15, 16, 16+. Panels A to C are presented on different y-axis scales for legibility.

Sources: 1940–2000 U.S. decennial census data and 2001–2020 American Community Survey data (IPUMS).

Heterogeneity by race/ethnicity and nativity

We find that the historical trends in educational homogamy identified in the aggregate—a decline between 1940 and 1960, then an increase until about 1990, followed by stagnation and decline—holds for the most part by race/ethnicity and nativity, but the levels of homogamy between groups differ considerably. Figure 4 presents trends in the odds of homogamy by race/ethnicity, and Figure 5 by nativity. The odds of homogamy are higher among Asian/Pacific Islander, Hispanic, and foreign-born women compared to White, Black, and native-born women. There is a clear decline in homogamy among native-born, White, and Black women since around 1990 or 2000. These results reveal that changes in population composition due to increasing immigration since 1970 have contributed to keeping educational homogamy high and relatively stable despite larger declines in the odds of homogamy in the native-born population. Future work can quantify the relative importance of these changes by developing a population decomposition approach to analyzing these trends. Like for the broader population, the increasing odds of hypogamy are an important component of the recent stagnation and decline in homogamy among these groups (see Online Appendix Figure S7).

Figure 4:

Odds of Homogamy, Different-Sex Prevailing Marriages by Wives’ Race/Ethnicity, 1940–2020.

Notes: Wives aged 18–40. Education categories are <10, 10–11, 12, 13–15, 16, 16+. Estimates use 3-year intervals between 2000–2020 for legibility.

Sources: 1940–2000 U.S. decennial census data and 2001–2020 American Community Survey data (IPUMS).

Figure 5:

Odds of Homogamy, Different-Sex Prevailing Marriages by Wives’ Nativity, 1940–2020.

Notes: Wives aged 18–40. Education categories are <10, 10–11, 12, 13–15, 16, 16+.

Sources: 1940–2000 U.S. decennial census data and 2001–2020 American Community Survey data (IPUMS).

Patterns among same- and different-sex married and cohabiting couples

Figure 6 shows the trends in the odds of homogamy for same- and different-sex married and cohabiting couples. First, comparing Figure 6 to Figure 1 shows that the results for different-sex couples are virtually identical when including cohabitors, which is not surprising given that cohabiting couples make up a small percentage of all coresidential couples. (Online Appendix Figure S8 shows that different-sex cohabiting couples have lower odds of homogamy than different-sex married couples.) Second, consistent with prior research, the odds of homogamy are lower among same- than different-sex couples and are also lower among male than female same-sex couples (Schwartz and Graf 2009; Verbakel and Kalmijn 2014), although other research has found a similar likelihood of homogamy for male same-sex couples and different-sex couples (Ciscato, Galichon, and Goussé 2020).⁹ The variability and time-span of our data are such that it is difficult to discern a clear trend for same-sex couples or any discontinuity around 2015 when same-sex marriage was legalized nationally. Future research should continue to track trends in assortative mating among same-sex couples as well as differentiate between married and cohabiting same-sex couples.

Figure 6:

Odds of Homogamy, Same- and Different-Sex Married and Cohabiting Couples.

Notes: Women aged 18–40 in different-sex couples and household heads aged 18–40 in same-sex couples. Unmarried and married co-residential couples are included for both different- and same-sex couples. Education categories are <10, 10–11, 12, 13–15, 16, 16+. Estimates for same-sex couples use 3-year intervals between 2008–2020 for legibility.

Sources: 1940–2000 U.S. decennial census data and 2001–2020 American Community Survey data (IPUMS).

Conclusion

Despite expectations for rising educational homogamy along with continued growth in economic inequality into the new millennium, we find that the degree of educational similarity between spouses stabilized beginning around 1990 and has declined since the 2000s. Increasing intermarriage across all educational boundaries and the rising prevalence of marriages in which wives have more education than their husbands contributed to this trend. We note that the recent pattern of declining educational homogamy is present within every population subgroup we examine among different-sex couples, suggesting broad-based social changes are at work. Partially offsetting these trends, however, is a compositional effect of the growing immigrant, Hispanic, and Asian/Pacific Islander populations for whom the likelihood of educational homogamy has long been higher than among native-born, White, and Black populations.

These results signal a reversal of a decades-long trend of growing social distance between educational groups in the United States. Given recent findings of reduced economic inequality since the early 2010s (Aeppli and Wilmers 2022), perhaps such a result would have been expected, however, the timing of the leveling and drop in educational homogamy occurred beginning in the 1990s and early 2000s—prior to reductions in economic inequality. In addition, reductions in economic inequality since the early 2010s have been concentrated in the bottom two-thirds of the income distribution and the odds of crossing all educational barriers have increased, even between college graduates and those with a graduate or professional degree. Thus, although future research should empirically investigate the mechanisms behind these trends, we note that the timing and pattern of change in assortative mating is more consistent with the reversal of the gender gap in education, which occurred in the population in these decades (DiPrete and Buchmann 2013; also see Appendix Figure S6).

Increasing intermarriage across educational boundaries suggests weakening social closure by education. Our results show that this has occurred across all educational barriers, but also that a major explanation is women’s increased likelihood of marrying less educated men. These findings shift the conversation about the potential impacts of assortative mating on inequality. Because women tend to earn less than men, the rise of wives with the educational advantage may drive the earnings resemblance between spouses further upward. Indeed, women who have more education than their husbands have annual earnings that are more similar to their husbands’ than women who have the same level of education (Schwartz and Han 2014: Table 2). Thus, increased hypogamy may be related to increased economic inequality across households at least in the short term (but see Boertien and Permanyer 2019).

Our results also point to the need to better understand couples in which women have more education than their male partners. How are these couples challenging or maintaining gendered narratives and expectations (Klesment and Van Bavel 2022; Qian 2017)? The implications of assortative mating for intergenerational inequality also hinge on whether the process of status transmission from one generation to the next differ depending on whether mothers or fathers have more education (Hu and Qian 2023; Jayet 2023). These are also relevant questions for the many countries around the world in which women have more education than men. We may expect similar declines in educational homogamy in these countries, given the rising prevalence of partnerships in which women have the educational advantage (Esteve et al. 2016).

Although our results point to the increased permeability of educational barriers as measured by educational intermarriage, it is notable that a college degree remains the least permeable of all the barriers we measure. Cross-education marriages are much more likely between those with a college degree and those with graduate or professional degrees, for instance, than between college graduates and those with “some college.” These findings are consistent with past scholars’ observations that a college degree is the key dividing line across which family patterns diverge (Lundberg et al. 2016).

Despite broad similarities in the overall trends among different-sex couples, we also show heterogeneity in the odds of homogamy by race, ethnicity, and nativity, and among same-versus different-sex couples. Past literature has investigated mechanisms that may contribute to differences between groups and change over time, albeit generally on a much more abbreviated time scale (e.g., Choi and Mare 2012; Lichter, Qian, and Song 2022; Thomas 2020). Future research should investigate causes of these differences as well as their consequences.