1Educational expansion and longer periods spent in school in France since the 1960s have strongly affected the ways couples form. Those changes partly explain the development of cohabitation outside marriage (Villeneuve-Gokalp, 1990) and the fact that individuals enter their first union at a later age (Galland, 1995; Prioux, 2003). The literature on partner choice has been highly attentive to these changes, focusing in particular on how they affect educational homogamy, or the tendency to choose a partner with the same or a similar level of education.
2Beginning with the observation of increasing educational homogamy in the United States (Mare, 1991; Schwartz and Mare, 2005), certain authors such as Blossfeld and Timm (2003) posited that lengthier education was the primary factor in a causal series that would ultimately exacerbate inequalities between households and may be transmitted from one generation to the next. According to this theory of ‘educational systems as marriage markets’, educational expansion bolsters social network homogeneity at ages when people typically form couples, thereby increasing the probability of meeting and forming a union with someone with an equivalent level of education. It holds that because educational attainment is a main determinant of an individual’s future occupational status and therefore of his or her income, polarization of economic resources and an opposition between advantaged and disadvantaged couples are inevitable.
3This apparently implacable logic nonetheless suffers from two major weaknesses. First, it is based on an extrapolation from findings of studies from the United States, whereas in Europe in the last decades, educational homogamy has fallen sharply in several countries, particularly France (for a review of the literature, see Bouchet-Valat, 2014). The second weakness concerns the supposedly mechanical reinforcement of income inequality between households caused by educational homogamy. Several studies have established that the ties between the two phenomena are quite weak (Worner, 2006; Breen and Salazar, 2010, 2011; Breen and Andersen, 2012; Eika et al., 2014; Courtioux and Lignon, 2015; Frémeaux and Lefranc, 2015), perhaps because educational homogamy is no guarantee that the partners’ careers will be more similar than those of the average couple a few years into the relationship.
4This article focuses on the connection between lengthened education and income inequalities. It analyses the context in which partners meet across generations, how that context interacts with educational homogamy, and career differences within couples. First, to what degree has educational expansion increased the proportion of couples formed during their years of study and, in particular, of partners who met at school or university? Second, are couples formed in this context more homogamous than others in terms of educational level and/or social background? Third, does homogamy at the start of the relationship have lasting effects on partners’ occupational positions despite career advancement inequality by gender?
5The EPIC survey on individual and conjugal trajectories (Étude des parcours individuels et conjugaux, INED–INSEE, 2013–2014), with its detailed questions on where partners meet and a life history approach making it possible to study partners’ occupational careers longitudinally, offers an unprecedented opportunity to examine these questions for France.
Review of the literature
1 – Meeting while in school and trends in places partners meet
6The sociological literature on couple formation has brought to light a change across birth cohorts in the types of places in which people meet their first partner. In France, Bozon and Héran (1989) found a diversification of meeting places and a sharp drop in the proportion of first meetings happening in the respondents’ neighbourhood between the early 20th century and the 1980s. In France in the 1960s, public dances (bals) or similar events were the most common places to meet a partner. They were then overtaken by nightclubs, the highpoint of which was in the 1980s, and parties among friends.
7The sharp rise in meeting at school is the most notable change since the 1960s. Whereas in that decade, 8% of men and 5% of women found their first partner at school, by the 1990s, the figures had risen to 18% and 15%, respectively (Bozon and Rault, 2012). However, school is hardly at the top of the list of partner meeting places, even if meetings occurring in other situations, such as evenings among friends, could be considered a feature of student socializing.
8Individuals still frequently meet their partners in ‘open places’. That is, following Bozon and Héran’s topology, they are meeting in places such as public dances (a bal or other public event, a nightclub) or other such venues (the neighbourhood, a public place, through a dating ad), where there is no admission selection. This trend has declined, though. In cohorts who formed couples between 1960–1968, such places accounted for 55% of meetings but only 34% in cohorts who formed a couple between 1991 and 1998 (Bozon and Rault, 2012). Similar trends have been found for Great Britain (Lampard, 2007).
2 – Does meeting in an educational context influence homogamy?
9It may seem inevitable that couples who meet at school are educationally homogamous, given that schools and universities are reserved to individuals selected on this very criterion. However, this does not mean that partners who meet in school or university classrooms necessarily attain the same level of education; for example, their education may be extended or prematurely interrupted. For older generations in particular, the roles of student and wife may have seemed incompatible given the dominant social norms (Singly, 1987). Furthermore, though the influential theory of educational systems as marriage markets attaches great importance to the link between meeting at school and educational homogamy (Blossfeld and Timm, 2003), there is little literature on that specific question.
10In France, Bozon and Héran observed that female higher education graduates who met their partners at university were slightly more likely than those who had met theirs elsewhere to form a couple with the son of an upper-level white-collar employee (cadre) (38% compared to 31%) and even more likely to do so with a future upper-level white-collar employee (60% compared to 43% for other meeting places). Meeting in an educational context therefore does seem to strengthen homogamy. Conversely, couples who meet in places that are atypical for their group are themselves atypical and therefore not strongly homogamous.  Offspring of upper-level white-collar workers who met their partners in an open place were half as likely to select other offspring of that same socioprofessional category. Conversely, offspring of blue-collar workers who met their partners in a closed place chose offspring of upper-level white-collar workers twice as often for females and 3 times as often for males.
11More recently, Bergström (2016) used EPIC survey data to show that, when controlling for union characteristics,  couples who meet on the Internet are less educationally homogamous than couples who meet at school or work. On the other hand, they do not differ in terms of occupational and social background homogamy.
12Some studies are available for countries other than France. One, from the United States, has shown that couples (whether married or cohabiting) who met at school or university are slightly more homogamous educationally than others (Laumann et al., 1994). Kalmijn and Flap (2001) have shown on the basis of Dutch data that when structural effects are controlled for in log-linear models, partners who attended the same school are more educationally homogamous than others. The other meeting contexts (work, associations, family, neighbourhood) do not have a statistically significant effect on educational homogamy, a finding that could be due in part to a small sample size. Nor is a significant effect found for social background homogamy. Finally, using loglinear models, Potarca (2017) has confirmed that meeting at school in the United States and Germany produces stronger educational homogamy than meeting anywhere else.
13While meeting at school therefore does seem associated with greater educational similarity, this finding has not been clearly established for France. In this article, we assess whether it applies to all birth cohorts for educational and social background homogamy.
3 – Partners’ careers: always better for men?
14Educational homogamy is not enough in and of itself to determine how similar partners’ occupational positions will be throughout their relationship, given that, even if they have the same level of education, their careers may follow quite different paths. These divergences are largely due to the gendered operation of the labour market and differences in career investment by gender.
15In the first place, women do not get as high of an occupational return on their education as men do (Mainguené and Martinelli, 2010). Secondly, women are more likely to work part time or to interrupt their careers, particularly after the birth of a child (Couppié and Épiphane, 2007; Meurs et al., 2010). Thus, even when controlling for the characteristics of the first job, the gender pay gap increases over men’s and women’s careers (Le Minez and Roux, 2001, 2002; Dupray and Moullet, 2005).
16These phenomena result in men having better careers than their women partners, i.e. a tendency towards female hypergamy in terms of occupational status or wage (Guichard-Claudic et al., 2009; Bouchet-Valat, 2017) that contrasts with women’s overall higher educational levels (Bouchet-Valat, 2015). The trend is only reversed when the woman’s educational level is markedly higher than her male partner’s or when his career runs into difficulties (Testenoire, 2008; Guichard-Claudic et al., 2009). The persistence of stereotypes that assign the private sphere to women and the work sphere to men helps perpetuate these gender behaviour norms, despite an overall receding of the notion that women are naturally less occupationally competent than men (Burricand and Grobon, 2015).
17Taking into account these inequalities within couples is essential to understanding partners’ respective careers and measuring change in occupational status differences (occupational heterogamy) over the relationship. Given that women’s occupational positions and therefore their incomes are on average lower than those of their male partners, even with the same level of education, the impact of educational homogamy on inequalities between couples may not be as strong as expected.
II – Data and methods
18As stated earlier, this article draws on data from the EPIC survey. A representative sample of individuals aged 26 to 65 living in metropolitan France was questioned about all of their ‘couple relationships or serious intimate relationships’ (relations de couple ou relations amoureuses importantes) either in the past or at the time of the survey.  The survey’s 7,825 respondents described a total of 14,699 relationships. EPIC is the only recent French source with detailed descriptions of places partners met and couples’ situations at the outset of their union (in particular, partners’ occupations); it is therefore well suited to a longitudinal study of homogamy.
1 – Definition of variables
19The relationships studied by the EPIC survey and throughout this article correspond to ‘couple relationships or serious intimate relationships’. This formulation encompasses a variety of unions without imposing a strict definition of a couple. Respondents were free to understand it in their own ways, which is likely to have produced varying interpretations. Younger respondents may attach more importance than older respondents to their first relationship, even if it did not last. This in turn could lead to overestimating increases in the proportion of first relationships formed in an educational setting. That is, there could be underestimation in older cohorts and overestimation in younger ones.
20No distinction is made here between cohabiting and non-cohabiting relationships or on whether unions have or had legal status. Non-cohabiting relationships account for 48% of first serious intimate relationships and 56% of current relationships that started less than 5 years before the survey, but for only 10% of all active relationships at that time. Because the focus here is on the place partners met, taking into account all types of relationships, without anticipating whether they will last, improves our view into relationship beginnings.  The results of the models used here do not change if we limit the study to cohabiting couples.
21To measure the frequency of couple formation during education, we begin by considering both first relationships and active relationships at the time of the survey. As educational level and social background were only collected for a respondent’s current partner at the time of the survey, we focus afterward on these current relationships. On average, these relationships are more stable than first relationships and therefore probably more homogamous because sharply heterogamous relationships are less stable (Mäenpäa and Jalovaara, 2014). Strictly speaking, there is no bias in proceeding this way: relationships in progress reflect the reality of unions at a given moment, and it is such relationships that determine inequalities between couples.
22The context of a couple’s meeting was collected through the following question: ‘Where did you first meet?’  Seven settings are distinguished: the Internet, a public place (a bal or other public event, fair, the neighbourhood, the street, a café, a store, etc.), a leisure venue (nightclub, discotheque, vacation spot, concert, association), an educational context (grammar school, middle school, high school, university), at work, with friends (parties with friends, private homes), through family (family gathering or party). In addition to the frequency of meeting in an educational setting, the frequency of relationships begun while the respondent was a student was determined by comparing the date the relationship began with the date marking the completion of the respondent’s education.
23Homogamy at the time the couple met was captured using two dimensions: the partners’ respective levels of education and their social backgrounds. Seven educational level categories were defined: no qualifications; general lower secondary qualification; vocational lower secondary qualification; upper secondary qualification; 2 years of higher education; 3 or 4 years of higher education; 5 or more years of higher education. Partners’ social backgrounds were approached through the father’s socioprofessional category: farmer; small entrepreneur; manager or professional; intermediate occupation; routine non-manual employee; manual labourer.  For each of these two dimensions, homogamy was defined as the two partners belonging to the same category.
24To study partners’ careers, the difference between their occupations was measured using a continuous indicator of prestige. We used the prestige scale developed by Chambaz, Maurin, and Torelli (1998) based on INSEE’s set of 16 socioprofessional categories.  The higher an occupation’s score, the more value society attaches to it. Scores were normalized to produce a spread from 0 to 100, ranging from unskilled workers to professionals.
25Two different indicators can be calculated using the prestige score. First, we use a heterogamy (the opposite of homogamy) indicator to measure the distance between partners. This assesses how distant the two partners’ occupational positions are from each other without noting whether it favours the man or the woman. Mathematically speaking, distance amounts to the absolute value of the difference between partners’ scores, ranging from 0 (for partners in equally prestigious occupations) to 100 (for an unskilled worker partnered with a professional). Second, we use an indicator of hypergamy to measure the gender gap; it is equal to the difference between the man’s and woman’s positions. Here we can measure whether men’s positions are higher than their partners’ or the inverse. This gap varies from −100 (for a male unskilled worker partnered with a female professional) to +100 (for a male professional partnered with a female unskilled worker); 0 is for partners whose occupations are equally prestigious.
2 – Coverage and methods of the longitudinal analysis
26The longitudinal analysis of change in the difference between occupational prestige scores within a couple is based on a subsample of relationships under way at the time of the survey in which each partner had already worked at the start of the relationship. This corresponds to 53% of relationships at the time of the survey. Couples formed when only one partner was at school were thus excluded.  Of this subsample of relationships, 34% were made up of at least one person who was economically inactive at the time of the survey. Those relationships were deliberately included, taking the last occupied job, the assumption being that they reflect the reality of interrupted careers, particularly those of women. Withdrawal from the labour market slows careers, and the last job held approximates (optimistically, no doubt) the position the individual might obtain if he or she were to return to work.
27Linear regression models were estimated on this subsample to grasp the mechanisms linking the gap between partners’ occupational statuses at the time of the survey (the dependent variable) to partners’ educational levels and the gap between their occupations when the partnership began (independent variables). After excluding couples in which one partner had never worked before the relationship began, non-responses for the independent variables came to 6%. Due to the lack of information regarding the partners’ occupations at the end of relationships that ended before the time of the survey, the effect of relationship duration can be biased by a selection according to the age at the beginning of the relationship or to the birth year (e.g. if less homogamous couples break up more frequently). This is why the relationship’s rank and the year of birth were added as control variables. Their effect was not significant, however.
28Restricting the analysis to people who reported having an occupation when their current relationship began may have induced a bias, though its meaning is hard to assess. As explained, the subsample excludes relationships that began during one partner’s education; that is, relationships that began at an early age or where both partners were in a lengthy education programme. To check the degree to which subsample selection could affect results, a regression model variant using Heckman’s two-step correction model was applied. The ages of the partners at the beginning of the relationship were used to predict whether the relationship was included in the subsample (that is, the fact that both partners had already worked at the beginning of the relationship). The test confirms the results obtained for the subsample.
III – Partners meeting more often in an educational context
29The first of the three claims made by the theory of educational systems as marriage markets posits that partners meet more frequently than before in educational settings or during formal education (Blossfeld and Timm, 2003). In this section, we test that statement by comparing first relationships to current relationships at the time of the survey.
30In France, educational expansion took place in two waves. The first comprised cohorts born in the early 1940s; the second concerned those born in the early 1970s (Chauvel, 1998). As a result, time spent in education increased more markedly than age at first union (Robert-Bobée and Mazuy, 2005; Rault and Régnier-Loilier, 2015). Consequently, the proportion of first relationships begun before the completion of formal education, as well as the proportion of partner meetings in the context of education, increased sharply across cohorts (Figure 1A). In cohorts born in the 1950s, fewer than 2 in 10 relationships began before the end of schooling (regardless of meeting place), and the figure rises to 5 in 10 for men born in 1980 and as high as 6 in 10 for women in the same cohort (women generally form couples earlier than men). Between the two cohorts, the proportion of partners meeting in an educational setting also rose, from 1 in 10 first unions to 2 in 10 for women and 3 in 10 for men. The lower frequency for women is due to their forming couples with men who, on average, are slightly older.  These tendencies seem to hold for younger cohorts, but people who had not been in a cohabiting relationship at the time of the survey might have defined their first important relationship more broadly than those who had. This would explain the apparent acceleration.
31Partner separations mitigate this trend. If we observe relationships under way at the time of the survey (only 45% of which were first relationships), the proportion of partners who meet in an educational context or before completing formal education rises across the cohorts, but much less sharply (Figure 1B). One in 10 people born in 1950 met the partner they were currently with at the time of the survey during their education, as opposed to 2 in 10 men and 3 in 10 women born in 1980. In the same cohorts, the proportion of meeting in an educational setting rises from slightly over 1 in 20 to 1 in 10. The increase over time might reflect only an age effect: the more recently the respondent was born, the younger he or she was at the time of the survey and the more likely to have experienced only one union.
Proportions of relationships begun during early education and of partner meetings in an educational context by birth cohort and sex
Proportions of relationships begun during early education and of partner meetings in an educational context by birth cohort and sexCoverage: Couples or ‘serious’ intimate relationships (individuals aged 26 to 65 living in metropolitan France).
32The rise in the proportion of first relationships begun in an educational setting is not as great as the first claim of Blossfeld and Timm’s (2003) theory seems to suggest. However, because the proportion of meetings during schooling increases over the generations, the question of the homogamy of couples formed in that context arises.
IV – Meeting in an educational context reinforces educational homogamy
33The second claim of the theory of educational systems as marriage markets is based on the assumption of stronger educational homogamy in couples whose members met at school. While this relation seems obvious, it has seldom been tested empirically. Here we also look at social background homogamy (partners’ fathers’ occupations), a variable that seems less obviously related to meeting in an educational context and so is interesting for the sake of comparison. The analysis in this section is restricted to active partnerships at the time of the survey.
34As expected, variations in homogamy by meeting context are quite wide (Figure 2), especially for educational homogamy, defined as the proportion of couples whose members have the same level of education (horizontal axis). Whereas only 35% of all couples are homogamous, the figure is 50% for couples whose members met in an educational context. In contrast, relationships that began on the Internet and in leisure venues are much less homogamous, at respectively 26% and 27%.  Between these two extremes, the figures for meeting through friends and family, work, or in a public place are close to the mean.
35Surprisingly, variations in social background homogamy (vertical axis) are quite different from those in educational homogamy. While couples whose members met at school are by far the most homogamous when it comes to educational level, they are the least homogamous for social background (28% compared to 33% of all partner meetings). In contrast, meetings in public places are associated with the strongest social background homogamy (39%), while their educational homogamy is close to the mean. Meeting in a leisure venue also results in weak educational homogamy but average social background homogamy. Contrary to what may have been supposed, then, meeting in an educational setting does not systematically favour homogamy.
Educational and social background homogamy at time of survey by meeting place
Educational and social background homogamy at time of survey by meeting placeInterpretation: For 39% of couples whose members met in a public place, the partners’ fathers belonged to the same socioprofessional category; the figure is 42% for individuals born between 1948 and 1967, and 35% for individuals born between 1968 and 1988. Meanwhile, the proportion of couples whose members have the same level of education fell from 38% to 31% between the two cohorts.
Note: The arrows indicate the direction of change between the two cohorts when it is statistically significant at the 5% level. Otherwise, only the mean point is indicated.
Coverage: Couples or ‘serious’ intimate relationships at the time of the survey (individuals aged 26 to 65 living in metropolitan France).
36Additional analyses were done using models that control for the structural effect of the social characteristics of individuals who met in each of these contexts.  First, social background homogamy is less strong in couples whose members met at school simply because that population is particularly heterogeneous. The children of upper-level white-collar workers (a small group in the sample) are over-represented, and manual and routine non-manual workers’ children (large groups) are under-represented: relative social background diversity mechanically increases the probability of individuals forming heterogamous couples. A similar level of homogamy would be found if the couples whose members met in this context had formed at random, through a random exchange of partners within the group. However, the stronger educational homogamy of partners who met as students cannot be explained solely by the composition of that population. The finding that partners’ educational levels are closer than they would be if all individuals who came together while students had paired off at random means that an additional selection takes place based on the level of education. These findings are consistent with those for other countries cited above (Kalmijn and Flap, 2001; Potarca, 2017).
37If we study change across the cohorts, we find that couples formed in an educational context are just as homogamous in the generations born between 1968 and 1988 as in those born between 1948 and 1967.  As a marriage market, then, the educational system does not favour homogamy after educational expansion any more than it did before.
38However, the homogamy rates for other meeting places have changed. First, homogamy has fallen sharply in couples whose members met in public places: from 38% to 31% for educational homogamy and from 42% to 35% for social background homogamy. This drop can be explained by the decline in the neighbourhood as a meeting place, whereas it played an essential role before, particularly in the lives of the working and agricultural classes (Bozon and Héran, 1989), working in favour of partners’ social proximity (educational homogamy rate of 56%, social background homogamy rate of 44% for the 1948–1967 cohort). The same phenomenon is found for couples meeting through family: 35% in the 1948–1967 cohort were homogamous for social background but only 26% in the 1968–1988 cohort. Indeed, no change is perceptible for educational homogamy.
39In contrast, 29% of couples whose members met at work were homogamous for social background in the first cohort, 35% in the younger one. This increase can be understood as a result of the improvement in women’s careers. An increase of women in the workplace is likely to increase the probability of socially equal individuals meeting. Indeed, the increase was found only for mid- and higher-level occupations.
40Finally, no statistically significant difference for female hypergamy—the tendency towards a gap in favour of males—was found by meeting context. Sample size was not high enough to determine whether the proportion of cases in which the woman had a higher educational level varied by partner meeting context (for couples whose members have different levels of education).
V – Effects of educational homogamy on career differences within couples
41In this section, we check the third and final claim of the theory of educational systems as marriage markets; that is, whether there is a link between educational homogamy (which does not vary after the relationship has begun) and similarity in partners’ occupational statuses. Does educational homogamy lead to more similar occupational statuses? If so, how does this fit together with the occupational gap between partners when couples form and with later career developments?
42To answer these questions, we first compare variations in occupational status differences between partners at the time of the survey with their level of education similarity. We then try to identify the mechanisms at work by taking a longitudinal approach that takes account of the initial occupational status difference. The occupation prestige scale mentioned in Section II was used to compute a distance between partners (heterogamy indicator, equal to the absolute value of the score difference) and a gender gap (hypergamy indicator, equal to the difference between the man’s and the woman’s occupational prestige scores).
1 – Partners’ occupational statuses are closer when they have the same level of education
43Whether calculated in terms of distance or gender gap, the occupational status gap at the time of the survey corresponds to an ending point resulting from both the gap at the time the couple formed and later developments in the partners’ respective careers. As predicted by Blossfeld and Timm (2003), similarity in educational level reduces occupational status distance (heterogamy indicator) when considering relationships of all lengths (Figure 3A). For partners with the same level of education, the occupational status gap at the time of the survey is 3 prestige points less than in couples where the woman has a higher level of education, and 7 to 8 points less than in couples where the man has a higher level (which amounts to distances, respectively, 14% and 30% below the corresponding reference results).  This observation holds across the generations.
Average occupational distances and gender gaps between partners at time of survey by birth cohort and educational homogamy
Average occupational distances and gender gaps between partners at time of survey by birth cohort and educational homogamyInterpretation: In cohorts born after 1968, the prestige distance between the occupations of partners (heterogamy) with the same levels of education attains a mean of 19 points (on a scale from 0 to 100) by the time of survey.
Coverage: Couples or ‘serious’ intimate relationships at the time of the survey where both partners had already worked when the relationship began (individuals aged 26 to 65).
44These effects differ sharply depending on whether the man or the woman is more educated, and analysis of the gender gap in occupational status finds and confirms considerable inequalities by sex (Figure 3B). Even when partners’ level of education is the same, the occupational gap is usually to the man’s advantage (hypergamy) at the time of the survey, in older and younger generations alike (9 and 7 prestige points, respectively). When the man has a higher level of education than that of his partner, it widens (20 points) without any observable difference by generation. It is when the woman has a higher level of education that the gender gap in occupation prestige at the time of the survey is narrowest. It is in the woman’s favour, on average—about 5 prestige points in post-1968 generations and only 1 point in older ones. While educational difference does affect occupational similarity within couples, occupation returns on education appear higher for men than for women, particularly in older generations.
45Analysing heterogamy and hypergamy indicators at the time of the survey confirms that occupational disparity between partners varies depending on their educational similarity, thereby seeming to validate the third postulate of the theory of education systems as marriage markets. However, the observed effects are limited. Moreover, we noted a tendency towards female hypergamy in terms of occupation, even when the partners had the same level of education at the beginning of the relationship, a finding that suggests the importance of studying how the partners’ careers evolved over time.
2 – Career differences between partners are not highly dependent on educational differences
46We now turn to the mechanisms that produce disparities between partners’ occupational positions at the time of the survey in order to identify the respective roles of educational homogamy and gender inequality over the course of a relationship.
47First, two linear regression models were estimated (Models A1 and A2 in Table 1) to explain heterogamy (non-gendered distance between partners) at the time of the survey. The analysis here confirms the effect of educational homogamy on occupational distance, with year of birth, union rank, partners’ employment statuses, and number of children born within the current relationship (Model A1). But the explanatory power of this type of homogamy is limited; it accounts for only 2% of variance (R2).  What is more, that power is just as weak when we model occupational distance at the beginning of the relationship rather than at the time of the survey, which confirms the limited influence of educational homogamy on the partners’ relative positions, regardless of the stage of the relationship. The small size of the effect is also confirmed: according to this model, and with other variables controlled for, if all couples were educationally homogamous (rather than the one-third of the sample that is), the average prestige distance between partners would only decrease from 23 to 20 points—a mere 13%. 
Results of linear regressions explaning occupational heterogamy and hypergamy at time of survey*
Results of linear regressions explaning occupational heterogamy and hypergamy at time of survey*Interpretation: In Model B2, the fact that the man has a higher level of education than the woman widens the gender gap in his favour at the time of the survey by 6.36 prestige points (on a scale from 0 to 100) compared to educationally homogamous couples. In the same model, a 1-point increase in occupational difference in favour of the man at the start of the relationship widens the gender gap by 0.71 points at the time of the survey for partners who have been together for less than 5 years.
Notes: Models A1 and A2 explain non-gendered distance between partners (heterogamy). Models B1 and B2 explain the gender gap in the man’s favour (female hypergamy). 95% confidence intervals indicated below the coefficients. Non-significant coefficients (p < 0.05) are in italics. Control variables not indicated in the table because their effects are not significant: respondent’s year of birth (Ref.: 1948), number of children (none, 1, 2, 3 or more), each partner’s employment status at the time of the survey (working, not working), rank of relationship (first, higher). First modalities are references.
* Centred variables (mean equal to 0).
Coverage: Couples or ‘serious’ intimate relationships at the time of the survey and in which both partners were working when the relationship began (individuals aged 26 to 65 living in metropolitan France); n = 2,817.
48As could be expected, initial occupational distance seems much more decisive in explaining occupational distance between partners at the time of the study: when this variable is added, the proportion of explained variance rises to 20% (Model A2). However, the influence of initial distance is also only partial: for couples who have been together for less than 5 years, 66% of initial distance is still found at the time of the survey. This effect decreases as the time that partners have been together increases (interaction effects range from −0.18 for 5 to 14 years, or a total effect of 48%, to −0.34 for over 35 years, or a total effect of 32%).  Occupational homogamy, then, is not given once and for all. It is important to note that even when occupational distance at the time the couple formed is controlled for, couples where the man has a higher level of education show at greater distance at the time of the survey (+7 points on average). This is explained by the gap’s tendency to widen in favour of the man, as is clear from the hypergamy models below.
49The same results are found in the hypergamy analysis; that is, the difference between the man’s occupation and the woman’s (Models B1 and B2). The quality of the model’s adjustment to the data improves considerably when gender inequalities are taken into account. For each model type, variance explained is nearly twice that obtained for the heterogamy results (R2 equals 12% for Model B1 and 17% with complete interaction between partners’ educational levels, rising to 35% if we integrate the gender gap at the start of the relationship; Model B2).
50In contrast to Models A1 and A2, the gendered model (B2) brings to light an effect of relationship length on hypergamy, though that effect is restricted to long-lasting couples. Partners who have been together for more than 25 years show a gap in the man’s favour 6 points wider than partners who have been together for less than 5 years. This result seems to indicate an increase in female hypergamy over the relationship years. However, it could also reflect stronger hypergamy of couples formed before 1990; that is, an effect of the period in which the relationship began, rather than its duration.
51An effect similar to the non-gendered model is found for the occupational gap: among partners who have been together for less than 5 years, 71% of the occupational gap in the man’s favour at the start of the relationship is likewise found at the time of the survey, after which the effect diminishes over time. In addition, the effect of educational level differences persists after controlling for initial gendered occupational gap: +6 points in favour of the man when his level of education is higher; −6 points in the opposite case (before controlling for the initial occupational gap, the respective effects were +12 and −11).
52Overall, then, the hypothesis that similarity of educational level will strongly affect economic inequalities between couples (Blossfeld and Timm, 2003) is largely invalidated for France. Educational homogamy itself only slightly determines occupational difference at the start of the relationship and at the time of the survey. It is true that the effect of educational difference between partners persists when we control for occupational difference at the start of the relationship, but both the explanatory power and the effect size are limited.
53Does longer education reinforce homogamy when partners meet and throughout their relationship, as predicted by the theory of educational systems as marriage markets (Blossfeld and Timm, 2003)? Recent and detailed data from the EPIC survey, which enable us to combine an analysis of where partners meet and homogamy with a longitudinal study of individual trajectories, suggest that that claim needs qualification.
54First of all, while the proportion of relationships in which partners met at school has increased for first unions—from 1 in 10 in cohorts born in 1950 to 1 in 4 in cohorts born in 1980—increased separation and repartnering rates considerably attenuate this change for all relationships. For these same cohorts, the proportion of relationships at the time of the survey that started at school rises from 1 in 20 to 1 in 10.
55Secondly, relationships begun in an educational context are characterized by greater educational homogamy: half of partners who met in this context have the same level of education, as opposed to a mean of one-third. However, the effect of this on homogamy among all couples—and especially on inequality between couples—is limited since such unions are still a minority. This finding is consistent with the observation that educational, social class, and social background homogamy have declined in France in recent decades (Vanderschelden, 2006; Bouchet-Valat, 2014).
56Thirdly, as the theory of educational systems as marriage markets holds, the distance between partners’ occupational positions is smaller—by 30%— when they have the same level of education than when the man’s is higher. But educational level has little effect on the occupational distance between partners at the time of the survey: strengthening educational homogamy would only bring partners’ occupational situations slightly closer and so would have little effect on income inequality between couples. Additionally, a gendered approach (distinguishing disparities in the man’s favour from those in the woman’s favour) better explains variation in the partners’ occupational positions within couples than an approach in terms of distance between partners, which cannot account well for major differences between men’s and women’s careers. Still, the fact that no information was available in the EPIC survey either on the first occupations of partners who met while students or on features of relationships before the one in progress at the time of the survey constitutes a limitation to this analysis.
57The findings here indirectly indicate that the major determinants of inequalities between households concern factors other than educational homogamy. They may have to do with economic factors, such as the recent fall in the male employment rate (Breen and Salazar, 2010; Bouchet-Valat, 2017) and rising high incomes (capital gains in particular; see Landais, 2008). Demographic factors are also important, particularly the decreasing proportion of individuals in relationships and the rise in single-parent families, a family type that now accounts for a greater proportion of poorer households (Bonnet et al., 2017) and which particularly concerns the least educated mothers (Acs et al., 2015). To improve understanding of change in economic inequalities between households, we need to be more attentive to those phenomena.
AcknowledgementsWe would like to thank Vianney Costemalle, Marie Bergström, Arnaud Régnier-Loilier, and Wilfried Rault, as well as the researchers in INED’s Fertility, Families and Couples research unit, the café seminar participants in INED’s Economic Demography research unit, and Population’s anonymous reviewers for their helpful and constructive comments.
‘Atypical’ (Bozon and Héran’s term) in the sense that members of their group rarely meet their future partners there.
Specifically, the age at which the relationship began, the year it began, the rank of the relationship in the respondent’s history, and the respondent’s socioprofessional category.
For more details on the EPIC survey, see Rault and Régnier-Loilier (2019) in this issue of Population.
The EPIC survey also takes this approach, collecting partners’ characteristics at the start of the relationship rather than at the beginning of cohabitation.
Slightly under 3% of individuals reported a second meeting place. Only the first was taken into account.
The question requested the father’s ‘main occupation’. In ambiguous cases, interviewers were instructed to request ‘the occupation the parent spent the most time in’. For non-responses (10% of cases), the mother’s occupation was used.
This scale, constructed from ratings on a prestige scale from 1 to 5 attributed to 122 occupational titles by a sample of individuals questioned by INSEE in France in 1996, reflects the ‘social evaluation of occupations’. It is categorized as a prestige scale as opposed to a socioeconomic position scale, which is based on objective criteria such as skill level or income. However, as the authors’ analyses show, those dimensions are implicated in the evaluation criteria respondents use.
The EPIC question explicitly excluded ‘summer jobs’.
On average and for all cohorts taken together, men’s first partners are 1.2 years younger than they are, and women’s first partners are 2.7 years older.
Relatively few couples met on the Internet (4% of current relationships), and 87% of them involved repartnering. If we control for age, the year the relationship began, and the relationship rank, the relatively low level of homogamy found for this meeting setting disappears (Bergström, 2016).
These log-linear models (here a log-multiplicative layer-effect model, or UNIDIFF) were used in several earlier studies (Kalmijn and Flap, 2001; Potarca, 2017). We have only summarized the main conclusions here for purposes of comparison. See Potarca (2017) for a more detailed presentation of the specifications. This approach based on meeting place data cannot be used to control for their attendance by people who did not form couples. Detailed results may be requested from the corresponding author; they are also available at http://bouchet-valat.site.ined.fr.
The proportions calculated for each cohort do not differ significantly; the 95% confidence intervals are, respectively, [43%, 59%] for the first cohort and [43%, 55%] for the second.
For reference, the prestige gap between mid-level teaching, health, and civil servant positions and higher-level civil servant and intellectual and creative professions is 10 points; the prestige gap between skilled and unskilled manual labourers is 27 points.
Or 7% if we integrate the 49 categories describing the full interaction between partners’ levels of education.
This prediction was obtained by subtracting the coefficients associated with ‘hypergamous’ and ‘hypogamous’ modalities times the proportion of the sample those groups represent (33% in both cases) from the mean observed distance.
Although the model controls for year of birth, it is impossible by definition to distinguish the effect of couple duration from the effect of the period in which the couple formed.