Opportunity and Change in Occupational Assortative Mating

Abstract

This article documents how opportunity and change in the U.S. occupational structure shaped patterns of occupational assortative mating between 1970 and 2015–2017. Trends in occupational assortative mating have often been cited as potentially contributing to the rise in economic inequality—the idea that doctors increasingly marry doctors instead of nurses—thereby exacerbating the concentration of resources among advantaged households. Previous estimates of trends in occupational assortative mating are now decades old and their impact on household inequality has not been quantified. Our results show large-scale change. The prevalence of dual professional couples nearly tripled between 1970 and 2015–2017. Changes were especially large among particular occupational combinations. For instance, male doctors have become increasingly likely to be married to female doctors, and male lawyers to female lawyers. Almost all of the changes in occupational assortative mating patterns, however, are accounted for by changes in the distributions of spouses’ occupations, for example, the rise of women in professional occupations. Because of this, the contribution of occupational assortative mating to the rise in economic inequality has been small. In the absence of any association between spouses’ occupations, observed increases in household earnings inequality would have been reduced by 5%. Although this is a small portion of overall changes in inequality, it is much larger than prior estimates of the effects of educational assortative mating on inequality, which recent studies have estimated to be essentially zero.

INTRODUCTION

Occupations are central features of stratification systems. For most people, they remain the primary determinant of economic standing (Mishel et al. 2012: Table 2.10), but occupations shape far more than economic resources. They confer social prestige and are associated with distinct lifestyles, tastes, attitudes, and beliefs (Alwin and Tufiş 2016; Bourdieu 1984; Treiman 1977; Weeden and Grusky 2012). Occupations may also affect opportunities for partnership formation given that they structure patterns of interaction in daily life and are associated with potential partners’ characteristics such as the number of hours they work, their job flexibility and autonomy, and travel responsibilities. These aspects of occupations are important factors in dayto-day aspects of romantic relationships and in mate selection, and thus may powerfully shape patterns of assortative mating (Hout 1982; Hunt 1940; Kalmijn 1994; Oppenheimer 1988).

Major shifts in the U.S. occupational structure may have had important consequences for assortative mating on occupation. One of the most dramatic changes has been the massive movement of women into the labor force. In 2016, women comprised almost half the labor force (47%) relative to just 38% in 1970 (Bureau of Labor Statistics 2017: Table 2). But change has been uneven. Women have increased their representation to a larger extent in professional occupations, whereas many other occupations remain strikingly gender segregated, for example, childcare workers and plumbers (Blau, Brummund, and Liu 2013; Cotter, Hermsen, and Vanneman 2004). At the same time, the decline of male-dominated blue-collar jobs has substantially changed the occupation distribution of potential male partners. Thus, the landscape of potential mates has shifted differently for men and women.

Table 2.

Percent Homogamous by Occupation Classification

Measure	Percent Homogamous (%)
	1970	2015–2017	Change
Observed
Big Class	25.2	33.4	8.2
Major Occupation	12.5	16.1	3.6
Microclass	3.7	7.6	3.9
Controlling for Changing Occupation	Distributions
Big Class	25.2	23.7	−1.6
Major Occupation	12.5	12.7	0.2
Microclass	3.7	4.9	1.2

Sources: 1970 U.S. decennial census and 2015–2017 American Community Survey. 1970 N = 329,698; 2015–2017 N = 1,029,573.

Although there is a voluminous literature on trends in educational assortative mating (for reviews see Blossfeld 2009; Schwartz 2013), there is very little work on occupational assortative mating. Most existing studies have used data from 1980 or earlier (Hout 1982; Hunt 1940; Kalmijn 1994). There are two studies of occupational assortative mating that use more recent data (Han and Qian 2021; Mansour and McKinnish 2018), but only one prior study has examined trends in occupational assortative mating, and this was using data from 1970 to 1980 (Kalmijn 1994). We use data from the U.S. decennial census and the American Community Survey from 1970 to 2017 to describe trends in occupational assortative mating, thereby updating the time series by more than thirty years.

In interpreting these trends, we draw on prior theory and research to test hypotheses about potential mechanisms that may explain trends in spouses’ occupational resemblance. Because of large increases in the representation of women in professional occupations, we would expect the prevalence of dual-professional couples to have increased because of increased opportunities for matching. But has the tendency for individuals to match on occupation—e.g., doctors marrying doctors—increased over and above what we would expect based on changes in availability? On the one hand, occupational matching may have increased beyond what we would expect based on changes in availability given that earnings inequality between occupations has increased and thus marrying across occupations may be more economically consequential (Fernández, Guner, and Knowles 2005; Mouw and Kalleberg 2010). By contrast, occupations may have less meaning in marriage markets today given that individuals change occupations to a greater extent over their lives than in the past (Hollister 2012), which would predict lower levels of spousal occupational resemblance. Our analyses bring over four decades of data to bear on these questions.

In addition, we quantify the extent to which changes in occupational assortative mating have contributed to increased household earnings inequality in the United States. Past research has focused on the contribution of educational assortative mating to inequality and has found little evidence that changes have resulted in increased economic inequality either in the United States or elsewhere (e.g., Boertien and Permanyer 2019; Breen and Salazar 2011). If changes in educational assortative mating have not increased economic inequality, have changes in occupational assortative mating? Occupational attainment is not synonymous with education, and earnings inequality between occupations has increased even after controlling growing inequality by education (Mouw and Kalleberg 2010). Thus, changes in occupational assortative mating may have had larger impacts on economic inequality than educational assortative mating. Other research suggests, however, that assortative mating may have little to do with the rise of the association between spouses’ earnings, instead pointing to increases in wives’ labor force participation as a key factor (Gonalons-Pons and Schwartz 2017). Again, however, this research has not addressed the role of occupations. Thus, the extent to which occupational assortative mating may account for increased earnings inequality across households is an open question.

Throughout our analyses, we use three different occupation classification schemes (big classes, major occupations, and microclasses) to test whether our results are sensitive to the aggregation of occupations, as is common in the stratification literature (e.g., Blau and Duncan 1967; Featherman and Hauser 1978). We focus on different-gender couples, although future work could extend these analyses to an examination of occupational matching among same-gender couples and across other important dimensions such as race, ethnicity, and nativity. In sensitivity analyses, we include cohabiting couples to mitigate concerns about the changing selectivity of marriage and compare our results to those estimated from samples of newly formed relationships. In addition, we widen the scope of previous analyses, which have focused only on dual-occupation couples (Han and Qian 2021; Hout 1982; Hunt 1940; Kalmijn 1994; Mansour and McKinnish 2018), by including all spouses, not just those who report occupations. This is particularly important given the large changes in women’s employment over this period.

INEQUALITY, ASSORTATIVE MATING, AND JOB STABILITY

Trends in the occupational resemblance of spouses have the potential for increasing economic inequality across households, but the impact of these trends depends on the size and direction of changes in observed patterns. Given large increases in the numbers of professional women in the U.S., we would expect growing numbers of dual-professional couples. These changes would be expected based on shifts in the distribution of occupations alone and we describe these and other observed patterns.

Also of interest is whether occupational assortative mating patterns have changed beyond what would be expected by shifts in men’s and women’s occupation distributions. Are the observed changes consistent with changes in availability or has a more fundamental reordering of matching patterns occurred since 1970? Theory and prior research from studies of inequality, assortative mating, and job stability point in different directions.

Some past research suggests an increased role of occupations in choosing a partner in the United States. One hypothesis is that, as economic differences between groups increase, the likelihood of “marrying down” will decrease given that it is increasingly economically costly to do so (Fernández et al. 2005). Because wage and earnings inequality across occupations have increased in the United States (Mouw and Kalleberg 2010; Weeden et al. 2007), this suggests that spouse’s occupational resemblance should have increased in turn. Note that we might also expect occupational resemblance to have increased merely because spouses’ educational resemblance has increased (Schwartz and Mare 2005). Inequality between occupations grew in the United States even after controlling for the shifting education distribution of workers in occupations and the growth of inequality by education (Mouw and Kalleberg 2010). Thus, increases in wage inequality by occupation are not merely a reflection of growing inequality by education. In addition, occupations may have become more salient in selecting a partner given that increases in age at marriage may mean that young people may be more likely to meet their partners at work or through work ties than in school (Mare 1991; Rosenfeld, Thomas, and Hausen 2019).

By contrast, evidence on the declining persistence with which individuals hold jobs within occupations suggests that occupations may play a declining role in the matching process. Among men, occupational switching has increased (e.g., job changes from mechanic to bartender) and switching between large occupational groups has also increased (e.g., job changes from professional to sales occupations) (Kambourov and Manovskii 2008). Similar forces have reduced job stability among women, but declining stability has been counteracted by women’s increased labor force participation leading to little change in stability over time for women (Hollister 2012). Given that individuals are changing occupations throughout their lives at an increased rate (at least for men), we might expect occupations to play less of a role in the matching process than they once did. Inequality within occupations has also increased, further reducing the predictive power of occupations as a measure of future economic success (Kim and Sakamoto 2008; Mouw and Kalleberg 2010).

Yet other evidence points to no expected change in occupational assortative mating. Weeden and Grusky (2012) found that large aggregates of occupations (big classes) have become less powerful predictors of lifestyles and political and social attitudes. They also found that the importance of occupations (microclasses) in predicting these individual characteristics has been relatively stable since the 1970s. Thus, if individuals are primarily sorting on occupations as markers of lifestyles, attitudes, and beliefs rather than on occupations as markers of economic success, patterns of occupational assortative mating may have remained constant. Regardless of which of these mechanisms is most consistent with observed patterns, our analysis adds to our understanding of the potentially changing role of occupations on another important dimension of inequality, who marries whom.

IMPLICATIONS FOR INEQUALITY

Although the hypothesis that increased spousal resemblance on socioeconomic traits (e.g., education, occupation, and earnings) will increase inequality between households given that it concentrates advantages and disadvantages within families is common and intuitive (e.g., Boertien and Permanyer 2019; Schwartz and Mare 2005), research on this topic has yielded mixed findings. Almost all previous research has found that the growing resemblance between spouses’ earnings has contributed to growing inequality in the United States (e.g., Gonalons-Pons and Schwartz 2017; Karoly and Burtless 1995; Schwartz 2010). By contrast, contrary to most scholars’ expectations, the bulk of the literature has found that changes in the resemblance between spouses’ education has had zero to negligible effects on inequality in the U.S. and Europe (e.g., Boertien and Permanyer 2019; Breen and Andersen 2012; Breen and Salazar 2011; Greenwood et al. 2014; Kremer 1997).

Breen and Salazar (2011) found that, rather than contributing to inequality, trends in educational assortative mating in the United States had a very small equalizing impact (1.4% reduction in inequality) between the late 1970s and early 2000s. Boertien and Permanyer (2019) similarly found an equalizing impact of 0.3% between 1974 and 2016. The main reasons given for these surprising finding have been that (1) the resemblance between spouse earnings is primarily the result of increases in women’s labor force participation, which have occurred for women at all education levels (Gonalons-Pons and Schwartz 2017) and (2) changes in educational assortative mating have not been large enough to produce sizeable changes in inequality (Boertien and Permanyer 2019; Kremer 1997).

Although research is ongoing in this area, at the very least, this literature suggests that the link between increased spousal educational resemblance and earnings inequality is not as straightforward as was once presumed. One link in the chain between education and earnings that has thus far been ignored is occupation. Occupation remains the key determinant of earnings in the United States (Mishel et al. 2012), and past research has found that increased spousal earnings resemblance is associated with increased inequality. Thus, if doctors increasingly marry doctors rather than nurses, for example, this will increase economic inequality across households, controlling for other changes. Another way that occupational assortative mating could increase inequality is via the large-scale shift from male breadwinner couples to dualearner couples (Ruggles 2015). Because spouses’ earnings are more similar when both partners work than if only one reports an occupation, this compositional shift alone could increase economic inequality across households (Schwartz 2010).

Changes in occupational assortative mating have the potential to produce larger effects on inequality than educational assortative mating given that earnings inequality between occupations has increased even after controlling for growing inequality by education. We use decomposition techniques to assess the extent to which trends in occupational assortative mating have contributed to earnings inequality across households in the United States since 1970.

DATA, MEASURES, & METHODS

Data

To describe trends in occupational assortative mating, we use data from the 1970, 1980, 1990, and 2000 U.S. decennial censuses and from the 2010–2017 American Community Survey (ACS) (Ruggles et al. 2010). Following Weeden and Grusky (2012), we exclude individuals living in group quarters and those with military occupations. Our primary sample is comprised of prevailing marriages in which wives are between the ages of 18 and 54; in other words, it is representative of approximately the working-age population of wives. The total sample size across all years is 4,270,463 couples. We often focus on end point years (1970 and 2015–2017) in our presentation of results for simplicity. Trends between these years are relatively monotonic.

To ensure that our results are not sensitive to our sample selection and to reduce concern about the effects of the changing selectivity of marriage, we replicated our analyses for newlyweds and cohabiting couples in the online appendix and discussed below. (See Schwartz and Mare 2005 for a discussion of the strengths and weaknesses of using data on newlyweds versus prevailing marriages).

Occupational Classification

We present trends and patterns using three occupational classification schemes (1) “big classes,” a 5-category occupational grouping, (2) “major occupations,” a 12-category grouping, and (3) “microclasses,” a 127-category grouping. We might expect trends in occupational assortative mating across these different classification schemes to vary in a way similar to Weeden and Grusky’s (2012) analysis of inequality by big classes and microclasses with stability among microclasses and a declining importance of big class. The primary goal of this article is to describe basic trends in occupational assortative mating using these different categorization schemes. Future research should explore how and why patterns may vary across these categories in more detail. We return to the topic of variation in trends across these three measures in the conclusion.

We define microclasses following Weeden and Grusky (2005, 2012) who demonstrate that microclasses are “good information-conveying ‘containers’” (2005:144) that differentiate between detailed occupations or small groupings of occupations defined by their distinct life chances, lifestyles, and attitudes. These categories are narrow and concrete enough to show occupations recognizable by laypersons (e.g., doctors, nurses, truckers, teachers), but are not so narrow to be unwieldy. The Weeden-Grusky scheme consists of 126 microclasses. Our scheme has an additional category for those who did not report an occupation. Respondents did not report an occupation if they had not been employed in the recent past. The “no occupation” group includes individuals who had not worked in the past 10 years in 1970 and those who had not worked in the past 5 years in 1980 and after. We use sex-specific weights to backcode detailed occupation codes to their 1970 counterparts and then group these occupations into microclasses (Weeden 2005; Weeden and Bucca 2016). This classification scheme has not been applied to couples before. The couple format of the data creates some unique backcoding challenges described in the proposed online appendix.¹

To summarize trends across larger groupings of occupations, we classify microclasses into big classes and major occupations. The big class categories are (1) professionals and managers (referred to henceforth as “professionals”), (2) sales and clerical, (3) crafts, operatives, laborers, and farm (referred to henceforth as “blue-collar” occupations for brevity), (4) service, and (5) no occupation. This categorization scheme roughly follows the aggregated groupings outlined by Featherman and Hauser (1978: 25–38)² as well as other classification schemes (e.g., Autor and Wasserman 2013).

We also use a 12-category major occupation classification similar to finer grained groupings used in prior research (Featherman and Hauser 1978; Torche 2011): (1) upper professionals, (2) lower professionals,³ (3) managers, (4) sales, (5) clerical, (6) craft, (7) operatives, (8) laborers, (9) service, (10) farm, (11) farm laborers, and (12) no occupation. For some analyses, we add a thirteenth major occupation group that identifies female-dominated professional occupations, which we define as microclasses in the professional category that were over 75% female in 2015–2017 (nurses and dental hygienists, therapists, health technicians, social workers, primary and secondary teachers, and librarians and curators). We do this because female-dominated professions differ from other occupations in ways that are likely to be associated with assortative mating (Levanon and Grusky 2016; McClintock 2020).

We examine trends in marriage patterns in more detail among upper professional and managerial occupations given that these are the occupations for which compensation rates have grown fastest over the past several decades (Cha and Weeden 2014) and, as we will demonstrate, changes in occupational assortative mating have been most pronounced. Strong perceptions of gender typical pairings also often characterize these occupations, e.g., doctors and nurses, managers and secretaries.

Methods

Our analysis proceeds in three steps. First, we describe trends in occupational assortative mating by big class, major occupation, and microclasses. Second, we show the extent to which these descriptive trends are a product of changes in spouses’ occupation distributions. Third, we estimate the extent to which changes in occupational assortative mating have contributed to trends in household inequality in the United States.

To describe trends in occupational assortative mating holding constant changes in couples’ occupational distributions, we use table-raking techniques (Agresti 1990:345–346, 643). Log-linear models are typically used in the assortative mating literature for this purpose, but results from table-raking methods are more readily interpretable and yield the same conclusions as log-linear models as assessed in sensitivity tests (not shown).

The table-raking procedure fits the model

$$\log \left(\frac{E_{ij}}{{\mu }_{ij}}\right)=\lambda +{\lambda }_i^H+{\lambda }_j^W$$ (1)

where H = husbands’ occupation category and W = wives’ occupation category, E_ij are the expected frequencies for the raked table, μ_ij = E(n_ij), and n_ij are the observed cell frequencies. The fitted values of the table can be expressed as:

$$\log {\widehat{E}}_{ij}-\log n_{ij}=\widehat{\lambda }+{\widehat{\lambda }}_i^H+{\widehat{\lambda }}_j^W$$ (2)

where −logn_ij is the adjustment term and is estimated using the log link function with an offset (Agresti 1990:346). Because, in some years, the data contain weights, our n_ij are the weighted cell frequencies (using the person-level weights of wives/female partners).⁴

To hold the marginal distributions of occupational categories constant at their 1970 values, we first compute pseudo-values of a contingency table with the observed time-varying marginal distributions of husbands’ and wives’ occupational categories for each year but in which the cell frequencies satisfy independence ([row total x column total]/n). This is logE_ij in equation (2). We then estimate equation (2) as a log linear model with an offset term (logn_ij in which the offset term is the weighted frequencies in each year. The expected frequencies from this model result in a new table with constant 1970 marginals but in which the associations (odds ratios) change as observed. A similar method was used by Breen and Salazar (2011: 828–9) in their analysis of the impact of increases in educational assortative mating on inequality in the United States.

To estimate the portion of the increase in household-level inequality that is due to occupational assortative mating we combine methods used by Cancian and Reed (1999) and Schwartz (2010). Schwartz (2010) examined the contribution of assortative mating to inequality among married couples; we extend this analysis to include all households using Cancian and Reed’s (1999:181) method, which divides inequality among households into subgroups of married-couple and other (e.g., cohabiting, single) households.

$$C{V}_h^2=S_m{(\frac{{\mu }_m}{\mu })}^{2}C{V}_m^2+S_B{(\frac{{\mu }_u}{\mu })}^{2}C{V}_u^2+\left[S_m{(\frac{{\mu }_m}{\mu })}^2+S_u{(\frac{{\mu }_u}{\mu })}^2\right]/{\mu }^2$$ (3)

where CV_ℎ² is the squared coefficient of variation among all households and is a common measure of inequality, S_m is the share of the population in married-couple households; μ is the overall mean earnings; μ_m is mean earnings for married-couple households; and $C{V}_m^2$ is the squared coefficient of variation for married-couple households. Quantities with “u” subscripts are comparable estimates for other households. We define married-couple households (group m in equation 3) as those in which both the householder and spouse’s marital status is “married spouse present” and the spouse is married to the householder. All other households are defined as group u in equation 3.

For comparability with our analysis sample of prevailing marriages, which is restricted to couples in which wives are between the ages of 18 and 54, our analysis of household earnings inequality is estimated from a sample of households in which the householder is between 18 and 54 years of age. We estimate the contribution of occupational assortative mating to household earnings inequality, where household earnings are measured as the sum of all household members’ wage and salary earnings. (In sensitivity tests, we adjusted for household size, but this did not substantially alter our results.) Because married couples represent a subset of all households, occupational assortative mating will necessarily have a smaller impact on household inequality than on married-couple inequality. Thus, in supplementary analyses, we also estimate the impact of occupational assortative mating on married-couple earnings inequality, where married couples’ earnings are measured as the sum of husbands’ and wives’ wage and salary earnings.

We modify equation 3, which gives observed trends, to estimate the impact of trends in occupational assortative mating on changes in earnings inequality. First note that occupational assortative mating is an occupation-level phenomenon. This means that any within-occupation contributions to increased earnings inequality are not due to occupational assortative mating, as we have defined it here. Thus, our first step in estimating the contribution of occupational assortative mating to increased inequality is to determine the portion of the change that is due to between-occupation variation and the portion due to within-occupation variation. We do this by replacing the squared coefficients of variation for married-couple households and other households in 1970 and 2015–2017 with the squared coefficients of variation of household-level occupational earnings (that is, the sum of household members’ median microclass earnings) letting all else vary as observed and re-estimating equation 3 for these years.

Next, we identify the contribution of occupational assortative mating to trends in inequality by employing two counterfactuals. The first counterfactual is: What would trends in inequality have been had the marginal distributions of husbands’ and wives’ microclasses changed as observed but the association between spouses’ microclasses remained as it was in 1970? The difference between the observed and counterfactual trends is an estimate of the contribution of changes in occupational assortative mating to increased inequality. To do this, we using table raking techniques (described above) that hold the association between husbands’ and wives’ microclasses constant at its 1970 values but allow the marginal distributions to vary. We then recompute the CV for married couples by year and replace the counterfactual CVs into equation 3.

The second counterfactual is: What would trends in inequality have been had the marginal distributions of husbands’ and wives’ microclasses changed as observed but spouses’ microclasses were independent? This counterfactual is similar to those used in other research that assumes random matching conditional on observed marginal distributions (e.g., Greenwood et al. 2014). We do this by constructing a counterfactual contingency table in which the marginal distributions change as observed but in which husbands’ and wives’ microclasses are independent from one another ([row total x column total]/n). The counterfactual trend in married couples’ CV is then estimated and used in equation 3.

While the first counterfactual assumes a constant association between spouses’ occupations, the second assumes constant independence. Thus, the difference between the second counterfactual and the observed trend is an estimate of the full contribution of the association between spouses’ microclasses to trends in inequality. One could also examine the difference between the first and second counterfactual trends, which is an estimate of what Schwartz (2010) refers to as compositional effects. To illustrate this concept with an example, because spouses’ occupational earnings associations are higher more when both partners work than if only one reports an occupation, and because the proportion of dual-occupation couples has increased, we would expect this compositional shift alone to increase inequality.⁵

It is important to note that our counterfactuals do not estimate the causal impact of what would have happened to all the many downstream outcomes if occupational assortative mating was, for example, random. Instead, they quantify the impact of the components of interest under well-defined assumptions, e.g., if the marginal distributions had changed as observed.

CHANGES IN THE U.S. OCCUPATIONAL STRUCTURE

Figure 1 shows well-known changes in the occupation distributions by gender, which provides the opportunity structure within which occupational assortative mating takes place. Key among these shifts is the massive entry of women into the labor market and the rise of women in professional occupations. Men’s representation in professional occupations has also increased since 1970 but less dramatically than women’s. Figure 1 also shows the large decline of men’s employment in blue-collar occupations, as well as the smaller decline of blue-collar occupations among women. The rapid rise of women in professional occupations compared with men’s blue-collar losses and their tepid increases in professional occupations has led some to observe that women have adapted to changing labor market conditions more successfully than men (Autor and Wasserman 2013). These structural shifts are associated with uneven change in the extent to which occupational groups are segregated by gender. The rise of women in professional occupations is associated with large declines in occupational segregation in these occupations, but other occupations have remained persistently gendered (Cotter et al. 2004; Levanon and Grusky 2016). These shifts are likely to have had substantial effects on patterns of occupational assortative mating.

Figure 1. Employment Shares by Major Occupation (Men and Women Aged 25–64).

Sources: 1970, 1980, 1990, and 2000 U.S. decennial census; 2010–2017 American Community Survey.

Notes: “Professionals” include professional and managerial occupations; “Blue Collar” includes craft, operative, laborer, and farm occupations.

OCCUPATIONAL ASSORTATIVE MATING BY BIG CLASS

The broad shifts in the occupational structure shown in Figure 1 are reflected in patterns of occupational assortative mating. Table 1 shows changes in the joint distribution of husbands’ and wives’ big classes between 1970 and 2015–2017. Table 1 is ranked by the percentage point change in couples’ relative prevalence in the population (Column C). Reflecting the increasing proportion of men and women in professional occupations, dual professional couples were the fastest growing group, increasing from 7% of all married couples in 1970 to 24% in 2015–2017, or more than tripled. Indeed, the top four fastest growing combinations all include a professional wife, mirroring the growth of women in the professions. Also notable is the decline of combinations prevalent in the 1970s, such as marriages between blue-collar husbands and wives with no reported occupation, blue-collar husbands and blue-collar wives, and blue-collar husbands and sales/clerical wives.

Table 1.

Percentage Distribution of Husbands' and Wives' Big Class

Wives	Husbands	Same big class?	1970 (%) (A)	2015-2017 (%) (B)	Change (C)	Change Controlling for Marginals (D)
Professionals	Professionals	Yes	6.76	23.92	17.16	−1.26
Professionals	Blue Collar	No	3.57	9.88	6.31	0.32
Professionals	Sales/Clerical	No	1.90	4.49	2.59	0.15
Professionals	Service	No	0.54	2.93	2.39	0.54
Service	Professionals	No	1.92	3.36	1.44	0.57
Professionals	No Occupation	No	0.06	1.13	1.07	0.24
No Occupation	No Occupation	Yes	0.38	1.20	0.82	−0.41
Service	Service	Yes	1.37	2.04	0.67	0.16
Sales/Clerical	No Occupation	No	0.11	0.74	0.63	1.16
Service	No Occupation	No	0.14	0.67	0.53	−0.85
Blue Collar	No Occupation	No	0.14	0.32	0.18	−0.13
Sales/Clerical	Service	No	1.84	1.83	−0.01	−0.05
No Occupation	Service	No	1.43	1.23	−0.19	0.02
Blue Collar	Professionals	No	1.48	1.24	−0.24	0.80
Service	Sales/Clerical	No	1.29	1.04	−0.25	0.27
Blue Collar	Service	No	0.91	0.46	−0.45	−0.67
No Occupation	Professionals	No	7.08	6.45	−0.63	0.48
Blue Collar	Sales/Clerical	No	1.25	0.38	−0.87	−0.02
No Occupation	Sales/Clerical	No	2.93	1.32	−1.61	0.09
Sales/Clerical	Professionals	No	10.29	8.49	−1.80	−0.60
Service	Blue Collar	No	8.27	6.18	−2.08	−0.15
Sales/Clerical	Sales/Clerical	Yes	5.80	2.73	−3.07	−0.50
Sales/Clerical	Blue Collar	No	14.75	7.94	−6.81	−0.01
Blue Collar	Blue Collar	Yes	10.91	3.50	−7.41	0.01
No Occupation	Blue Collar	No	14.88	6.53	−8.35	−0.18
Total			100.00	100.00

Sources: 1970 U.S. decennial census and 2015–2017 American Community Survey. 1970 N = 329,698; 2015–2017 N = 1,029,573.

Notes: "Professionals" include professionals and managers; "Blue Collar" includes craft, operative, laborer, and farm occupations. Bolded rows indicate couples who share big class.

In addition, the bolded rows of Table 1 show changes in the percentage of couples sharing the same big class—that is, big class homogamy. Column C shows that, in general, the percentage of couples sharing the same big class increased among groups that grew (professionals and service occupations) and declined among those that shrank (sales/clerical and blue-collar occupations). These patterns are not surprising given changes in the occupational structure, but they nevertheless represent major changes in the composition of different-sex spouses’ occupations.

Increases in homogamy for some groups and decreases for others resulted in a small overall increase in the proportion of all couples who share the same big class—an increase in big class homogamy from 25% to 33% between 1970 and 2015–2017 (Table 2). This is a small change for a period of over four decades. We return to the question of the extent to which these patterns are altered when shifts in husbands’ and wives’ occupation distributions are controlled later in this article.

THE RISE OF WOMEN IN PROFESSIONAL OCCUPATIONS

As is evident in Table 1, a potential implication of the increasing share of women in professional occupations is the rise of dual professional couples. Table 3 shows that across all twelve major occupation groups, men were far more likely to be married to professional women in 2015–2017 than in 1970. For example, 58% of upper professional men were married to professional women in 2015–2017 up from 27% in 1970. Similarly, only 6% of operatives were married to professional women in 1970 and this increased more than four-fold to 28% in 2015–2017.

Table 3.

Marriage to Professional Women by Husbands' Major Occupation

	Percent (%) Married to Professional Wife		Of Husbands Married to Professional Wives in 2015-17, Percent (%) Married to…a
Husbands' Major Occupation	1970	2015-2017	Change	Upper Professional	Lower Professional	Female-Dominated Professional Occupation
Upper Professional	27.4	58.4	31.0	37.4	18.9	43.7
Lower Professional	31.9	60.8	28.9	14.8	20.8	64.4
Managers	18.4	50.4	32.0	18.8	24.1	57.1
Sales	16.0	48.6	32.5	17.0	24.6	58.5
Clerical	12.8	40.7	27.8	16.5	21.8	61.7
Craft	7.7	32.2	24.6	11.5	20.3	68.2
Operatives	5.9	27.6	21.7	11.4	20.0	68.6
Laborers	5.2	24.7	19.6	11.5	20.2	68.3
Service	9.0	34.5	25.6	12.2	18.2	69.6
Farm	9.8	36.8	27.0	12.9	16.7	70.4
Farm Laborer	4.3	17.3	13.0	11.4	19.6	69.0
No Occupation	6.9	27.8	20.9	20.7	19.6	59.7

Sources:1970 U.S. decennial census and 2015–2017 American Community Survey. 1970 N = 329,698; 2015–2017 N = 1,029,573.

Notes:

^aCategories are mutually exclusive. Female-dominated professional occupations are defined as those that were over 75% female in 2015–17 (nurses and dental hygienists, therapists, health technicians, social workers, primary and secondary teachers, and librarians and curators).

Yet men in different positions in the occupational distribution are married to different types of professional women. Table 3 shows that in 2015–2017, upper professional men (e.g., lawyers, physicians and dentists, engineers) were much more likely than other men to be married to upper professional women. Men in management and sales occupations were somewhat more likely than other men to be married to lower professional women (e.g., accountants, authors and journalists, public relations professionals) and blue-collar men were more likely than other men to be married to professional women in female-dominated professional occupations (such as nurses, teachers, and social workers), which tend to have lower earnings than other professional occupations (Levanon, England, and Allison 2009). Thus, men across each of these occupational groups became more likely to be married to professional women, but the types of professional women to whom they were married varied considerably.

The rise of women in professional occupations may mean that professional women face increased competition on the marriage market for professional men. For instance, in 1970, female doctors represented only 9% of all doctors whereas they comprised 40% of doctors in 2017 (authors’ calculations from Census and ACS data). Table 4 shows the percentage of couples that are homogamous by major occupation groups, conditional on husbands’ or wives’ occupation. For example, among husbands who were upper professionals in 1970, only about 4% were married to wives who were also upper professionals. By contrast, among upper professional women in 1970, a much larger percentage (33%) were married to upper professional men, likely reflecting the greater availability of professional men relative to women at that time.

Table 4.

Percent Homogamous and Spouses' Occupational Earnings by Major Occupation and for Upper Professionals

	Conditional on Husbands' Occupation			Conditional on Wives' Occupation
Measure & Group	1970	2015–2017	Change	1970	2015–2017	Change
Percent Homogamous (%)
Major Occupation
Upper Professional	3.7	17.2	13.5	33.4	38.1	4.6
Lower Professional	27.9	40.9	13.1	20.0	16.0	−4.0
Managers	4.3	17.3	13.0	26.1	29.4	3.3
Sales	9.3	9.8	0.5	10.5	13.0	2.5
Clerical	36.8	24.1	−12.7	9.0	6.0	−3.0
Craft	1.7	1.7	0.0	31.4	24.8	−6.6
Operatives	20.9	7.8	−13.0	34.3	29.8	−4.5
Laborers	1.4	3.3	1.9	9.7	15.6	5.9
Service	22.5	24.0	1.5	10.5	15.3	4.8
Farm	6.0	8.0	1.9	67.6	43.1	−24.5
Farm Laborer	14.7	18.2	3.5	35.8	37.3	1.5
No Occupation	45.7	29.6	−16.1	1.4	7.2	5.8
Upper Professional Microclasses
Architects	0.5	5.0	4.5	20.0	13.3	−6.7
Engineers	0.2	4.1	3.9	17.9	15.0	−2.9
Natural Scientists	2.2	9.8	7.6	12.0	13.1	1.0
Engineering & Science Tech	0.6	1.5	1.0	4.1	5.0	0.9
Doctors	2.6	18.4	15.8	36.5	29.7	−6.8
Other Health Professionals	2.5	14.5	11.9	7.0	8.9	1.8
Professors & Instructors	9.7	15.5	5.8	26.5	15.2	−11.3
Lawyers	0.8	12.6	11.9	21.9	19.5	−2.4
Computer Specialists	2.6	7.7	5.1	12.3	25.7	13.4
Spouses' Average Occupational Earnings (2016 $s)
Major Occupation
Upper Professional	38,609	50,535	11,926	56,204	57,415	1,212
Lower Professional	38,922	46,544	7,622	52,472	48,434	−4,038
Managers	36,902	45,223	8,321	50,716	47,999	−2,717
Sales	36,472	44,844	8,372	48,136	46,522	−1,614
Clerical	35,485	41,386	5,901	48,855	43,652	−5,203
Craft	34,271	38,799	4,529	45,098	42,237	−2,860
Operatives	33,718	37,321	3,603	43,466	37,412	−6,054
Laborers	32,547	36,427	3,880	43,039	39,301	−3,738
Service	33,330	38,388	5,059	43,413	39,107	−4,307
Farm	33,915	40,003	6,089	30,617	36,729	6,112
Farm Laborer	29,782	32,298	2,515	33,009	30,330	−2,679
No Occupation	33,324	39,263	5,939	47,371	44,932	−2,439
Upper Professional Microclasses
Architects	37,897	52,976	15,079	74,204	61,849	−12,355
Engineers	35,565	50,433	14,868	57,250	60,073	2,823
Natural Scientists	38,119	54,501	16,382	62,662	60,913	−1,749
Engineering & Science Tech	33,170	42,636	9,466	56,423	50,580	−5,842
Doctors	38,548	76,331	37,783	66,467	89,142	22,675
Other Health Professionals	36,929	54,502	17,573	56,431	60,701	4,270
Professors & Instructors	42,652	52,145	9,493	62,042	56,995	−5,047
Lawyers	39,628	59,487	19,859	64,562	67,438	2,876
Computer Specialists	36,427	50,860	14,433	60,384	60,232	−152

Sources: 1970 U.S. decennial census and 2015–2017 American Community Survey. 1970 N = 329,698; 2015–2017 N = 1,029,573.

Note:s Occupational earnings are calculated as the median wage and salary income in an occupation group for prime aged workers (aged 25 to 54) in a given year and are inflation adjusted to 2016 dollars using the CPI-U. "Doctors" include physicians and dentists.

Focusing on changes in homogamy by gender and major occupation between 1970 and 2015–2017, Table 4 reveals that increases in occupational homogamy have been concentrated among men with high status occupations. The growth of major occupation homogamy was substantially larger for professional and managerial men than for men in all other major occupation groups. From the perspective of women, homogamy by major occupation did not grow nearly as much as it did for high status men. The likelihood that upper professional women were married to upper professional men increased by only 5 percentage points whereas this likelihood increased by 14 percentage points for upper professional men. Moreover, the likelihood of homogamy for lower professional women declined by 4 percentage points whereas it increased by 13 percentage points for lower professional men.

Table 4 shows similar results for microclasses within the upper professional group. For almost all of these microclasses, e.g., doctors, architects, and lawyers, husbands saw their microclass homogamy increase more than did wives. For example, the percentage of male doctors married to female doctors increased by 16 percentage points (a more than 600% increase) whereas the comparable percentage for female doctors declined. That is, male doctors are much more likely to be married to female doctors, but female doctors are less likely to be married to male doctors. The results are similar for lawyers and other upper professionals. These results are consistent with the massive increase in the availability of female professionals and the resulting decline in the relative availability of male professionals.

What are the implications of these trends for spouses’ occupational earnings? Table 4 shows average spouses’ occupational earnings by wives’ (husbands’) major occupation. Occupational earnings are defined as the median wage and salary income in an occupation group for prime aged workers in a given year in constant 2016 dollars and each spouse in the data has an occupational earnings score. For example, upper professional husbands in 1970 had wives with occupational earnings that averaged $38,609. Focusing on changes in spouses’ occupational earnings conditional on wives’ major occupation, Table 4 shows that husbands’ occupational earnings decreased across almost all of wives’ major occupation groups with the exception of upper professional women and women with a farm occupation.⁶ The decline of husbands’ occupational earnings by wives’ occupation is consistent with stagnation in the median earnings of men working in full-time, year-round jobs and with increases in the proportion of men who are unemployed or out of the labor force (Mishel et al. 2012:Figure 5L; Semega, Fontenot, and Kollar 2017:Figure 2).

Figure 5. Changes in Occupational Homogamy among Upper Professional Husbands Controlling for Changing Occupational Distributions: 1970 and 2015–2017.

Notes: Percentages shown are conditional on husbands’ microclass and hold the marginal distributions constant at their 1970 values.

Sources: 1970 U.S. decennial census and 2015–2017 American Community Survey.

Figure 2. Changes in Wives’ Occupation from 1970 to 2015–2017 Among Male Doctors, Managers, Lawyers, and Engineers.

Sources: 1970 U.S. decennial census and 2015–2017 American Community Survey.

Notes: “Doctor” includes physicians and dentists; “Teachers” are primary and secondary school teachers.

Turning to microclasses among upper professional women, it is evident that there is considerable variation in the gains and losses associated with husbands’ occupational earnings. The rise in husbands’ occupational earnings among female doctors is particularly large. Despite the decline in microclass homogamy among female doctors, their husbands’ average annual occupational earnings increased by more than $22,000 in 2016 dollars since 1970. This did not occur because female doctors were married to husbands in different occupations with higher occupational earnings. Rather, female doctors’ husbands’ occupational earnings increased primarily because a substantial proportion of female doctors remained married to male doctors in 2015–2017, and the occupational earnings of doctors increased substantially over this period. Thus, the proportion of female doctors married to male doctors declined, but the returns to doing so increased. Husbands’ occupational earnings would have declined for female upper professionals on average if not for the increasing occupational earnings of doctors.

Similarly, female lawyers were less likely to be married to male lawyers over this period but retained roughly similar husbands’ occupational earnings. Again, this is because the returns to marriage to male lawyers increased and offset the increasing likelihood that female lawyers were married to men with lower occupational earnings. Thus, other than women married to men in a few elite occupations in which compensation increased over this period, the occupational earnings of husbands declined on average.

DECLINES IN GENDER TYPICAL PAIRINGS

Media accounts of changes in occupational assortative mating often feature changes in stereotypically gendered pairings such as doctors and nurses or managers and secretaries (e.g., Bennhold 2012; Miller and Bui 2016). The results above suggest that male doctors are increasingly likely to be married to female doctors. Are they also less likely to be married to female nurses? Similarly, are male managers and lawyers less likely to be married to female secretaries? This section focuses on trends in marriage for four historically male high-status professions—doctors, managers, lawyers, and engineers—and describes changes in their marriage patterns since 1970.

Figure 2 shows the large increase in occupational homogamy among male doctors also seen in Table 4. In addition, it shows that male doctors are indeed somewhat less likely to be married to female nurses than in the past (10% in 1970 versus 8% in 2015–2017). In general, male doctors have become more likely to be married to professional women and less likely to be married to women with lower occupational prestige (women in blue-collar, sales, clerical, and service occupations). They have also become less likely to be married to women who did not report an occupation.

The stereotypical pairing of managers and secretaries has also become less common. Male managers have become much less likely to be married to secretaries (14% in 1970 versus 5% in 2015–2017) and are instead married to other managers and professionals. Trends in marriage patterns among male lawyers are similar—male lawyers are much more likely to be married to female lawyers, other professionals, and managers, and are much less likely to be married to secretaries and women who did not report an occupation than they were in 1970.

A substantial fraction of male engineers in 1970 were married to primary and secondary school teachers as well as women in non-professional occupations. Virtually no male engineers were married to female engineers. Like other occupations, male engineers are now more likely to be married to female engineers, but the growth of this group is not as large as for doctors, managers, and lawyers. Engineering is still a highly male dominated occupation—80% of engineers were men in 2017 (authors’ calculations from Census and ACS data). Although a large proportion of male engineers were still married to primary and secondary school teachers in 2015–2017, they were also much more likely to be married to women holding other professional and managerial occupations than in the past.

SIMILARITY IN OCCUPATIONAL EARNINGS

Increases in wives’ occupational earnings and declines in husbands’ occupational earnings (Table 4) suggest an increasing resemblance between spouses’ occupational earnings. In addition, the disproportionate rise in homogamy among male professionals suggests that increasing similarity may be larger among this group than other occupations. Figure 3 shows that this is indeed the case. The correlation between spouses’ occupational earnings increased for all couples regardless of husbands’ big class category, but more so professional husbands. The correlation for couples with professional husbands was essentially zero in 1970 but rose to 0.22 in 2015–2017. The rise for husbands in other occupational groups was less extreme. These results are consistent with earlier findings that the resemblance between spouses’ occupational earnings increased between 1970 and 1980 (Kalmijn 1994), and indicate that this trend has continued. We quantify the extent to which these trends have contributed to increasing inequality in the U.S.

Figure 3. Correlation Between Spouses’ Occupational Earnings, Overall and by Husbands’ Big Class.

Sources: 1970, 1980, 1990, and 2000 U.S. decennial census; 2010–2017 American Community Survey.

Notes: “Professionals” include professional and managerial occupations; “Blue Collar” includes craft, operative, laborer, and farm occupations.

TRENDS CONTROLLING FOR SPOUSES’ CHANGING OCCUPATIONAL DISTRIBUTIONS

As we have shown, patterns of occupational assortative mating have changed markedly since 1970. The United States has shifted from a country in which different-gender marriages were often comprised of blue-collar husbands and stay-at-home wives to one with large numbers of dual professional couples. The rise of women’s occupational status has meant that men of all occupations are more likely to be married to professional women. These trends, however, may largely be driven by changes in the occupational structure, in particular, the increasing numbers of professional women. To what extent do the patterns we have described hold when changes in spouses’ occupational distributions are controlled? The answer to this question sheds light on theory and research about the direction of change in occupational assortative mating in the United States.

Column D in Table 1 shows changes in the distribution of spouses’ big classes holding constant changes in occupation distributions at their 1970 values but allowing the associations to change as observed using table-raking techniques.⁷ The primary conclusion from these results is that changes in patterns of occupational assortative mating are overwhelmingly the result of changes in the distributions of spouses’ occupations. Holding spouses’ occupational distributions constant vastly reduces observed changes across all categories and completely explains the increasing prevalence of dual professional couples. The prevalence of dual professional couples would have declined by about one percentage point over this period had the distributions of spouses’ occupations remained constant and the association changed as observed. By contrast, the likelihood that professional women were married to non-professional men increased somewhat across all husbands’ major occupational groups. This is consistent with the hypothesis that normative pressures for women to avoid “marrying down” have declined (Esteve et al. 2016). Controlling for the marginal distributions, homogamy by big class overall declined by 1.6 percentage points (Table 2).

Perhaps changes in occupational assortative mating patterns controlling for the marginal distributions only appear small because we have obscured important variation using the big class classification scheme. However, we also see small changes controlling for spouses’ changing occupational distributions using the 12-category occupation classification. Figure 4 shows cell percentages of couples on the diagonal of a cross-classification of spouses’ major occupations, holding the marginals constant at their 1970 values. It shows that the small decline in homogamy controlling for changes in the big class occupation distributions for professionals shown in Table 1 (Column D) is the result of small declines among lower professionals and managers rather than upper professionals. Across each major occupation group, the changes over four decades are small and thus should be interpreted as more stability than change. Overall, the percentage of couples sharing the same major occupation once changes in the marginals are accounted for increased by just 0.2 percentage points since 1970 (Table 2).

Figure 4. Percentage Homogamous by Major Occupation Controlling for Changing Occupational Distributions: 1970 and 2015–2017.

Sources: 1970 U.S. decennial census and 2015–2017 American Community Survey.

Note: Percentages shown are cell percentages for diagonal cells of a cross-classification table of husbands’ by wives’ major occupation category, holding the marginal distributions constant at their 1970 values.

Is there more evidence of change in occupational assortative mating patterns by microclasses controlling for the marginal distributions? Figure 5 shows homogamy trends for upper professionals, conditional on husbands’ microclasses and controlling for shifts in the distribution of spouses’ microclasses. It shows that, if the distributions of husbands’ and wives’ microclasses had remained at their 1970 levels but the association between spouses’ microclasses had changed as observed, male architects, doctors, lawyers, and computer specialists would have been less likely to be married to same-occupation women in 2015–2017 compared with 1970, but natural scientists, engineers, and other health professional husbands would be more likely to be married homogamously. Compared to the much larger increases in observed homogamy trends for male upper professionals shown in Table 4, these changes are very small.

Across all microclasses, cross-cutting trends results in a small increase in the percentage of couples sharing the same microclass controlling for shifts in the distribution of spouses’ microclasses of 1.2 percentage points (Table 2). Nevertheless, because overall levels of homogamy by microclass are low, this amounts to a 32% increase from 1970, but over a 45-year period this is not a large change.⁸ These results are consistent with Kalmijn’s (1994) findings, showing an increase in homogamy using a 70-category occupational grouping between 1970 and 1980, and indicate a continuation of this trend, although the changes we observe are small.

Across all of the occupational classification schemes, Table 2 shows that observed increases in homogamy are substantially lessened and small once changes in the marginal distributions are controlled.

THE RISE OF EDUCATIONAL HOMOGAMY & TRENDS IN OCCUPATIONAL ASSORTATIVE MATING

The relative stability of occupational homogamy controlling for the marginals is inconsistent with the increasing resemblance of spouses’ education (Schwartz and Mare 2005), although other research has shown less evidence of a trend (Rosenfeld 2008). Given that occupational position is strongly associated with education, we would expect trends in educational assortative mating to be similar to those for occupation. However, it is also possible that, as occupations become more unstable and occupational earnings more variable, individuals have increasingly shifted to relying on education rather than occupation as a marker of lifestyle and long-run earnings in their search for partners on the marriage market.

Supplementary analyses presented in the online appendix replicate and extend Schwartz and Mare’s (2005) analysis of trends in educational homogamy into 2015–2017, adding the dimension of occupational assortative mating. We find that our conclusions about trends in occupational assortative mating are not altered when trends in the association between education and occupation and trends in educational assortative mating are controlled. This suggest that trends in occupational assortative mating are relatively independent from trends in educational assortative mating and do not follow the same trajectories, a finding that is reminiscent of earlier research showing sorting on occupational earnings increased from 1970 to 1980 but decreased on occupational education (Kalmijn 1994). Future research should investigate the relationship between educational assortative mating and occupational assortative mating more closely.

NEWLYWEDS & COHABITORS

Trends using prevailing marriage data reflect not only patterns of sorting into marriage, but selective martial dissolution, occupational changes within marriage, and changes in the selectivity of marriage. We can come closer to measuring sorting into marriage by examining newlyweds and address changes in selection into marriage by including cohabiting couples. We replicated our analyses using a combined sample of newlyweds and cohabitors in which female partners are between the ages of 18 and 54 (results shown in online appendix). The results are very similar to those for prevailing marriages. Like for prevailing marriages, dual-professional couples have become the most prevalent couple type among newlyweds and cohabiting couples, while the share with a blue-collar partners declined precipitously. Big class, major occupation, and microclass homogamy all increased for newlyweds and cohabitors as they did for prevailing marriages, but these changes were smaller because a larger fraction of newlywed and cohabiting women reported an occupation in 1970 and thus were more likely to share an occupation with their male partners or husbands. Mirroring the pattern for prevailing marriages, controlling for the marginal distributions erases increases in big class and major occupation homogamy since 1970 and also the rise of dual-professional couples, but a small increase in microclass homogamy remains. That the results for newlyweds and cohabitors are so similar to those for prevailing marriages speaks to how powerful the structural economic changes have been that shape patterns of occupational assortative mating.

IMPACT ON INEQUALITY

How have changes in occupational assortative mating affected trends in earnings inequality in the United States? Given that, for example, male doctors are increasingly married to female doctors, we would expect the association between spouses’ earnings to have increased and inequality to have risen in turn.

Row (1) of Table 5 shows observed trends in occupational earnings inequality. Between 1970 and 2015–2017, household earnings inequality for this population increased by 49% (0.336/0.692*100). Part of this is because of increased inequality between occupations and part is due to increased inequality within occupations (Mouw and Kalleberg 2010). By replacing individuals’ earnings with their microclass occupational earnings and re-estimating changes in the CV (row 2), we estimate that 61% of the increase in household earnings inequality is due to within-microclass variation in earnings ((0.336–0.132)/0.336*100). This portion of the trend in household inequality cannot be explained by occupational earnings inequality or occupational assortative mating as it is due to within microclass variation.

Table 5.

Decomposition of Change in Earnings Inequality from 1970 to 2015–2017

	1970	2015–17	Change
(1) Total Observed Change in Household Earnings Inequality (CV)	0.692	1.028	0.336
Counterfactual Changes in Household Earnings Inequality (CV)
(2) No within-microclass variation in earnings	0.279	0.410	0.132
(3) Marginal distributions vary, occupational assortative mating held constant at 1970 values	0.279	0.412	0.133
(4) Marginal distributions vary, spouses' occupations independent	0.265	0.379	0.114
Of Change in Household Earnings Inequality:
% Due to within microclass earnings variation among married couples [(1)-(2) / (1)*100]			60.90
% Due to Changes in OAM [(2)-(3)/(1)*100]			−0.39
% Due to OAM [(2)-(4)/(1)*100]			5.22

Sources: 1970 U.S. decennial census and 2015–2017 American Community Survey.

Notes: Household earnings inequality is estimated for all households in which the householder is between the ages of 18 and 54 1970 N = 391,568; 2015–2017 N = 1,810,677

Next, we identify the contribution of occupational assortative mating to trends in the CV. Rows (3) and (4) of Table 5 show two counterfactual trends. Row 3 shows the counterfactual: What would trends in inequality have been had the marginal distributions of husbands’ and wives’ microclasses changed as observed but the association between spouses’ microclasses remained as it was in 1970? The difference between the observed and counterfactual trend is an estimate of the contribution of changes in occupational assortative mating to increased inequality. Given that our earlier results showed little change in patterns of occupational assortative mating controlling for the marginals, it is unsurprising that the percentage of observed change in household earnings inequality that is due to changes in occupational assortative mating is so small as to be negligible (−0.39%).

Row 4 shows the counterfactual: What would trends in inequality have been had the marginal distributions of husbands’ and wives’ microclasses changed as observed but spouses’ microclasses were independent? The difference between the second counterfactual and the observed trend is an estimate of the full contribution of the association between spouses’ microclasses to trends in inequality. The total impact of occupational assortative mating patterns to trends in household earnings inequality is larger but is still small, explaining about 5% of the increase in household earnings inequality. That is, if spouses’ microclass categories in 1970 and 2015–2017 were independent of each other conditional on the observed marginal distributions, household earnings inequality would have risen 5% less than observed. This result indicates that all of the small impact of occupational assortative mating patterns on inequality is the result of the movement of couples from areas of table in which the association between spouses’ occupational-earnings is low (e.g., doctors and nurses or doctors and stay-at-home wives) to areas in which it is higher (e.g., doctors and doctors). In a parallel analysis of the contribution of occupational assortative mating to increased inequality among married couples (rather than all households), we estimate a contribution of about 10% (not shown)--still relatively small but twice as large as the impact for households.

DISCUSSION

Our study has documented large-scale change in the occupational composition of marriages. In 1970, marriages between blue-collar husbands and stay-at-home wives and dual blue-collar marriages were common. In the 2010s, dual professional couples were the most prevalent couple type by a large margin. In fact, men of all occupational groups have become more likely to be married to professional women reflecting the growing number of professional women in the population, but men with less prestigious occupations are more likely to be married to women in female-dominated lower paying professions such as teachers and nurses whereas higher status men are more likely to be married to higher status female professionals such as lawyers and doctors.

We also show that the oft-told story of rising occupational homogamy applies almost exclusively to men in high status occupations. Occupational homogamy grew substantially for professional and managerial men, but not for other men or for women. Past research has not uncovered this because of its focus on gender-symmetric patterns of association controlling for marginal distributions (Hout 1982; Kalmijn 1994). Indeed, as the numbers of professional women have grown and the relative availability of professional men declined, occupational homogamy among a number of high status occupations held by women declined. For example, female doctors, lawyers, professors, and engineers were less likely to have same-occupation husbands in 2015–2017 than in 1970. By contrast, occupational homogamy for men in each of these occupations increased, and for some, notably doctors and lawyers, quite substantially. Occupational homogamy for men outside professional and managerial occupations either declined or increased to a much smaller extent, consistent with large and persistent occupational sex segregation in many of these occupations.

Declines in middle- and lower-skilled men’s earnings and small but steady declines in men’s employment are reflected in declines in husbands’ occupational earnings among wives in almost all occupation groups. A notable exception is upper professional women for which husbands’ occupational earnings have remained steady since 1970 mostly because of the outsized earnings growth of a few occupations, doctors and lawyers in particular. Declines in men’s occupational earnings combined with increases in women’s have increased the correlation between spouses’ occupational earnings for men in all big classes, although again this trend was especially pronounced for professional husbands.

The large shifts in patterns of occupational assortative mating we document are largely accounted for by changes in the U.S. occupational structure. There have been small changes in the percentage of couples sharing the same occupation controlling for the changing distribution of spouses’ occupations. This is true for big classes, major occupations, and microclass groups. The largest percentage change occurred among microclasses, for which homogamy increased from 3.7% to 4.9% between 1970 and 2015–2017, or by just 32% over a more than 45-year period. Supplementary analyses showed that trends are broadly similar for newlyweds and cohabiting couples.

Future research could use longitudinal data to determine how patterns of occupational assortative mating unfold across the life course and also better understand how couples meet. Prior research shows that meeting through friends was the most common way that different-gender couples met over this period, and the proportion of couples who met at work or through coworkers has declined since the early 2000s while meeting online and in bars or restaurants has increased (Rosenfeld et al. 2019). There is a class dimension to meeting as well, with middleclass couples more likely to meet through closer friends and family and working-class couples more likely to meet through weaker ties (Sassler and Miller 2014) and prior research has found that field of study in college plays a role in shaping occupational homogamy as does the gender composition of occupations (Han and Qian 2021; Mansour and McKinnish 2018), but we lack detail on how patterns of meeting vary by occupation.

What are the implications of our findings? First, our results suggest that the increasing prevalence of dual professional couples is entirely driven by changes in the distributions of spouses’ occupations, notably the increase in female professionals. There is no increase in dual-professional marriages over and above what we would expect based on the changing occupational distributions alone. Second, we might have expected more of an increase in occupational homogamy given large increases inequality and the rising costs of “marrying down” (Fernández et al. 2005). For example, big class homogamy only increased from 25% to 33% from 1970 to 2015–2017 and not at all once we controlled for the shifting distributions of spouses’ occupations. It is possible that increases in job instability, occupational switching across the life course, and increased earnings inequality within occupations (Hollister 2011; Sakamoto and Wang 2020) has reduced the power of occupations as a predictor of future economic standing in mate selection. Individuals may instead increasingly rely on more fixed characteristics such as education in selecting a spouse. Counterbalancing this downward pressure on homogamy may be the continued relevance of occupations for lifestyles, attitudes, and beliefs (Weeden and Grusky 2012) and the fact that occupations have become more consequential in explaining variation in wages and earnings (Mouw and Kalleberg 2010). These countervailing forces may have resulted in the small changes we observe.

Because trends in occupational assortative mating holding constant changes in spouses’ occupation distributions were small, we found that they had virtually no impact on trends in household earnings inequality since 1970. Occupational assortative mating did result in a small increase in household inequality via changes in the spouses’ occupational distributions. Because the earnings resemblance of spouses is higher among professionals and couples in which both partners work for pay, the growth of dual professional and dual occupation couples resulted in a small increase in earnings inequality. We estimate that if there was no occupational assortative mating (different-gender spouses’ occupations were independent of one another) but the distributions of these spouses’ occupations changed as observed, increases in earnings inequality across households would have been about 5% lower than observed. This is a small contribution to inequality, and we interpret our estimate in the context of many other factors associated with greater inequality including the decline of manufacturing jobs, unions, and the real value of the minimum wage, rising demand for skill, increased automation and globalization, and other changes in the family, in particular, the rise of single-parent families (Autor 2014; McLanahan and Percheski 2008). One reason why occupational assortative mating does not have a large impact on household inequality is because a substantial portion of trends in household inequality are not attributable to changes in occupational earnings; we found that 61% of the increase in inequality was attributable to within microclass inequality trends between 1970 and 2015–2017 and thus is not explainable by assortative mating at the microclass level.

To put the magnitude of our 5% estimate in context, most scholars describe the contribution of the rise of single-parent families to inequality as moderate or sizeable and these estimates range from 11% to 41% (McLanahan and Percheski 2008). Restricting the sample to just married couples increases the contribution of assortative mating to inequality trends to 10%, a larger but still small impact. These findings suggest that the larger impact of increased earnings associations that Schwartz (2010) and others have found on increases in earnings inequality is likely due to increased earnings associations within microclasses rather than between microclasses. Nevertheless, a 5% contribution to household inequality represents a substantially larger impact than recent studies of educational assortative mating have found. Recent studies of the impact of educational assortative mating have found very small inequality ameliorating effects but these are small enough to be effectively zero (Boertien and Permanyer 2019; Breen and Salazar 2011). Thus, examining a type of assortative mating that is a step closer to earnings in the chain from education to occupation to earnings yields substantially larger (albeit still small) estimates of the impact of assortative mating on trends in household earnings inequality.

Our results add another dimension to recent research about changing forms of inequality in the United States. Weeden and Grusky (2012) found that big class inequality in the United States has declined since the early 1970s. Big classes no longer predict lifestyles or political and social attitudes as well as they once did, although they do still predict differences in income. Consistent with this, we found a decline in big class homogamy controlling for spouses’ changing occupational distributions, although our big class categories are coarser than theirs. By contrast, Weeden and Grusky found microclasses have remained strong and consistent predictors of lifestyles, life chances, and attitudes. Once changes in the distribution of spouses’ occupations are accounted for, we also found relative stability in microclass homogamy. Unlike big classes, this suggests that microclasses are still good “information containers” not only in terms of lifestyle and life chances, but also in terms of marital sorting.