1In France today, women are more highly educated than men. This has affected first union formation and the traditional tendency – known as female hypergamy – for women to form unions with men who are more educated than themselves. Using data from the Family History survey (Étude de l’histoire familiale, INSEE-INED, 1999), Milan Bouchet-Valat analyses the educational characteristics of partners in a first union and the rate of permanent singlehood by educational level across cohorts, starting with those born in the 1920s. Disentangling structural effects from gender effects, he questions the existence of a norm of female hypergamy capable of withstanding the changing gender composition of the highly educated, and reveals deep-seated changes in the prevalence of male and female permanent singlehood by educational level.
2In addition to the well-known tendency to choose a partner with similar social characteristics, known as assortative mating (for a literature review, see Bouchet-Valat, 2014a), the strongly gendered nature of heterosexual partner choice has long been observed by anthropologists and sociologists. In many societies, this behaviour takes the form of female hypergamy or male hypogamy, i.e. a propensity for individuals to form couples in which the woman is of lower status than the man in one or more respects : age, height, stratum of a caste system, educational level and occupation, or social status more generally. The family being a central institution of socialization and social reproduction, one may suspect that constructing a position of inferiority for women represents a key factor of male dominance and its persistence across generations. So it is interesting to examine how female hypergamy has been affected by the major changes in women’s status over the twentieth century.
3This article focuses on educational hypergamy as an indicator of partners’ social status at first union formation, education being the main determinant of individuals’ lifestyle and peer associations at this age (Pan Ké Shon, 1998 ; Erlich, 1998), but also one of the strongest indicators of their future working career and social status. The level of education is one of the key items of information available to individuals – consciously or unconsciously – when they meet potential partners. On the contrary, jobs occupied over the working career may partly reflect choices made after union formation, making it difficult to clearly discern the gendered criteria of partner choice.
4Moreover, educational level is the dimension of social status that has changed most radically in terms of gender disparities. The educational expansion that occurred in France after the Second World War affected both sexes, but girls especially (Baudelot and Establet, 1992). Now that women are more highly educated, on average, than men, couples where the woman is more academically qualified than her partner have inevitably become more frequent than the reverse (Guichard-Claudic et al., 2009) – unless, that is, the most qualified women remain single. In this respect, women’s educational superiority is on a different level from the more limited and localized changes concerning, for example, the gender division of labour, and constitutes a case of large-scale “gender reversal” (Kergoat et al., 2008). If women are to develop the potential to overtake their partners in their careers, then a high cultural capital, even if not always used in working life – or not fully at least – is an essential prerequisite. Alongside other major transformations, such as the massive growth in female labour force participation, this movement may contribute to a change in the balance of power within the couple, and hence a questioning of gender roles – even if it probably lacks the strength to produce a result of this kind on its own.
5Recent research (Esteve et al., 2012) has shown that while female educational hypergamy was more frequent than the reverse in most countries of the world in the 1970s, it has declined substantially since them, to the point of becoming a minority situation in the 2000s in many societies. These findings suggest a need to revise an outdated conception of the relationship between gender, education and union formation whereby a high level of education is a handicap for women on the marriage market. This article looks more closely at this shift in behaviours by examining changes across cohorts in France.
6We begin by describing the signification of female hypergamy and the mechanisms that explain its existence, the changes one might expect to see across cohorts, and the results of previous research conducted in various countries. Using data from the Family History Survey (Étude de l’histoire familiale, EHF, INSEE-INED, 1999), we will then study the first unions of cohorts born between 1920 and 1970, exploring trends in hypergamy first in terms of raw composition, then controlling for the influence of constraints arising from the educational distribution of male and female populations within each cohort (relative hypergamy) as a means to capture changes in social norms and individual preferences. Last, we will analyse changes in permanent singlehood by sex and education using an accelerated failure time model to reveal the link between gendered distribution of singlehood and hypergamy. The article concludes with a discussion of the uncertain impact of the observed changes on gender inequality in other areas.
I – Theoretical perspectives and current knowledge
The classic concept of female hypergamy : a strongly gendered model
7Female hypergamy can only be understood in the context of a male-dominated society, being both a consequence of this norm and a vehicle for its perpetuation. In a model where women are generally in a position of inferiority, the reproduction of the social order requires that male dominance be respected within the family, the central institution for the interiorization of gender roles (Goffman, 1977). Under this rationale, men choose wives of lower social status than themselves, and vice-versa. Such behaviours reflect the interiorization of dominant norms (Bozon, 1991), the desire to avoid social disapproval – either explicit or implicit – or simply to forestall the negative feedback to which couples who violate these norms are exposed (Lefeuvre, 2008 ; Testenoire, 2008). Male superiority within the couple also ensures that, in most social situations, the roles assigned to man and woman are spontaneously assumed, thereby consolidating gender stereotypes in the face of realities that may challenge them. This is the case, for example, for the allocation of physical tasks to men, given that the man is generally taller (and hence, by convention, stronger) than his spouse more frequently than pure chance would dictate (Goffman, 1977 ; Herpin, 2003). Female hypergamy also provides grounds for assigning domestic tasks to the woman (Ponthieux and Schreiber, 2006), and for sacrificing the woman’s career, judged less promising, in favour of the man’s (Nicole-Drancourt, 1989 ; de Singly, 1987). Conversely, if the woman’s wage is higher than the man’s, management of household income is more egalitarian (Testenoire, 2008).
8The norm of hypergamy has powerful consequences for singlehood, placing certain groups – the most educated women and the least educated men – in a less favourable position on the marriage market. Among the French prewar cohorts, a high educational level had opposite effects for men and women, increasing the chance of finding a partner for the former, and decreasing it for the latter (Robert-Bobée and Mazuy, 2005 ; Winkler-Dworak and Toulemon, 2007) The highest educated men were less likely than might be expected to choose a woman of the same educational level as themselves, showing a preference for (slightly) less educated women, while the men just below them behaved likewise. This pattern was perpetuated all the way down the social scale. Given that the least educated men and the highest educated women were too socially distant to form unions with each other, and that the man’s visible inferiority with respect to his wife would have been unacceptable to both of them, the people in these two extreme groups were therefore condemned to singlehood.
9It is the singlehood of highly educated women that interests us the most here, since the gender norm at its origin runs directly counter to the rationales of social reproduction whereby high-status individuals tend to prefer partners with the same social standing. Perhaps under this model it is social origin that defines a woman’s social status, rather than her educational level. But this does not explain why a higher educational level has a negative effect on a woman’s chances of finding a life partner. Two mechanisms may account for this phenomenon.
10First, as stated by François de Singly (1987, p. 176), the source of singlehood among highly educated women may arise from “a poor match between the timing of educational dowry formation and the timing of arrival on the marriage market”. As remaining in education was deemed to be incompatible with union formation, especially before the advent of modern contraception, educated women were not available on the marriage market until a relatively late age. Men therefore preferred to pair up with women who had little education but who were no longer in school – given that choosing an older, highly qualified, woman violated the norm of female hypergamy in terms of both age and cultural capital. This mechanism made it very difficult for educated single women above a certain age to find a partner.
11A second factor concerns the desirability of a union, from the viewpoints of both the women concerned and of their potential spouse. On the one hand, educated women with more to lose on the labour market were less willing than other women to sacrifice their career for the sake of a constricted family life or an inegalitarian marriage (Cacouault, 1984 ; Kaufmann, 1999) ; these relatively high expectations were bolstered by the financial independence afforded by a working career. In the eyes of men, on the other hand, a high qualification could be seen as a signal of disinterest in family life, a lack of femininity and an independence that ran counter to their idea of married life (de Singly, 1987).
12In this highly gendered model, female hypergamy thus has three components. First, among couples that are already formed, the man more often has a higher level of education than the woman. Second, under the norm of female hypergamy, this situation persists beyond the constraints of partner availability (relative hypergamy). Last, alongside this majority of couples formed under the female hypergamy norm, the strongly gendered workings of the marriage market produce a large number of outcasts – educated women and uneducated men – at opposite ends of the social scale.
Changes over time : factors behind the weakening of hypergamy
13Over the course of the twentieth century, women’s status progressively began to catch up with that of men, as evidenced by their rising educational level, their mass participation in the labour force, the development of contraception, and the end of men’s guardianship over their wives. This change has radically altered the strongly gendered model described above. How has it affected female hypergamy ?
14As women have become more educated, hypergamy should inevitably have fallen from favour. Considering all the above-mentioned changes, one would expect to observe a decline in relative terms (i.e. after controlling for changes in the educational distribution of the population), and a reduction in permanent singlehood among highly educated women. Indeed, given the sheer scale of the changes under way, they cannot have occurred without a radical shake-up of gender norms, and without a broader knock-on effect on this population group.
15One can therefore expect men’s criteria for choosing a partner to more closely resemble those once favoured by women, now that highly qualified women with a working career are no longer seen as deviant or lacking in femininity. In this respect, changes in norms, behaviours and social reproduction strategies are converging. Indeed, with the expansion of education, academic achievement has become a key factor of social stratification in all social groups. Given that it is the mother who most frequently helps with her children’s homework (Héran, 1994), her educational level is very important for their school career (Place and Vincent, 2009). Without assuming that individuals apply an implacable logic to their romantic choices, it is reasonable to imagine that this new situation is contributing to a change in how men perceive the qualities of a potential partner, and to a weakening of the belief that education is incompatible with family life.
16Moreover, in purely economic terms, while a woman’s education could be seen as a negative signal for the prewar generations, this is clearly no longer the case. By making men’s working careers more uncertain, the rise in unemployment has made it very risky to adopt the strategy of Becker’s model (1981) whereby it is in the man’s interest to specialize in the labour market, and in the woman’s to specialize in the domestic sphere to ensure higher rates of return. The woman’s job, even if less well paid and highly precarious, can provide an appreciable safety net if the man finds himself out of work (Oppenheimer, 1994). This argument is especially pertinent in the French case, where the cohorts born since the 1960s have been exposed to high unemployment, but where women’s careers, while still less gratifying than those of men, are nonetheless better than in many other western countries (Kempeneers and Lelièvre, 1993).
17But changes in patterns of hypergamy are not solely the result of changes in male preferences and strategies : women’s preferences on the marriage market may also have moved closer to those of men. Their labour income gives them greater freedom to wait for a partner who corresponds to their wishes before committing themselves to a relationship. Less dependent on their future partner’s career, they can focus on his other qualities. This is revealed in the adjectives – less often centred on work and protection, and more often on feelings – that they now use to describe him (Bozon, 1991).
18Things are not that clearcut, however. While female labour force participation surged between the first and last cohorts in our study, and while the very nature of female employment has changed since the postwar period, gender equality in terms of wages and career advancement is still a distant prospect (Maruani, 2000), and women still earn much less than their partner (Morin, 2014). So one might rightly expect to observe a weakening of female hypergamy, but a weakening that is only partial, after controlling for changes in the educational distribution of the population. This hypothesis would imply that social norms are showing a degree of resistance – or of inertia at least – to the changes under way, because the model in which the man’s social status counts more than the woman’s still lives on. It does not seem to be supported by existing research, however.
Existing research : decline and reversal of hypergamy
19Most recent studies on this question have observed a decrease, and often even a reversal of educational hypergamy. In a broad comparative analysis covering 56 countries on all continents and at all levels of development, Albert Esteve, Joan García-Román and Iñaki Permanyer (2012) observed a very clear shift. Among the 18 countries for which the authors had data from the early 1970s, female hypergamy was always more prevalent than hypogamy among young cohabiting couples at that time. But by the early 2000s, the situation was reversed in 26 of the 51 countries included in the sample at that date. In a more detailed working paper, the same authors (Esteve et al., 2011) studied relative hypergamy, i.e. after controlling for the educational distribution of male and female populations. No apparent divergence between absolute and relative hypergamy is observed : partner choice thus reinforces female hypergamy in societies where women are generally less educated than men, and hypogamy in those where they are more educated. In other words, the behaviour of individuals always seems to reflect and amplify the constraints imposed upon them by their social environment. This finding appears to question the existence of a hypergamy norm : nowhere does such a norm seem to have withstood the changes in educational distribution in favour of women.
20These developments are confirmed by several other studies. David Monaghan (2014) observed that relative female hypogamy is stronger than relative hypergamy in 13 of the 26 developed countries studied between 1980 and 2010 (average of the observed years, young cohabiting couples). Several studies of individual countries have found similar tendencies. In the United States, for example, Elaina Rose (2004) notes that educational hypergamy has practically disappeared, and Zhenchao Qian (1998), followed by Christine Schwartz and Robert Mare (2005), even measured a reversal in absolute and relative terms. The same pattern was observed by Albert Esteve and Clara Cortina (2006) in Spain, and by Elina Mäenpää and Marika Jalovaara (2014) in Finland, and several studies in South America have obtained similar findings (Esteve and McCaa, 2007 ; López-Ruiz et al., 2009 ; Rodríguez, 2014).
21Last, concerning France, using data from the Family and Employers survey (Familles et employeurs, INED, 2004-2005) Yvonne Guichard-Claudic, Armelle Testenoire and Danièle Trancart (2009) found that educational hypogamy was almost twice as frequent among women aged 25-30 as among those aged 40-50. Among these cohorts born between 1955 and 1980, female educational hypergamy is always less frequent than hypogamy – although the reverse is true in terms of occupational category. Last, in a study of cohabiting couples aged 30-60 based on data from the French Labour Force surveys, we observed that female educational hypergamy became less frequent than hypogamy in around the year 2000 (Bouchet-Valat, 2014a).
22Contrary to what is suggested by the theoretical considerations presented above, the norm of female hypergamy – measured via relative hypergamy – seems to have reversed in many societies. Unfortunately, trends in the level of permanent singlehood by educational level, indissociable from hypergamy, have rarely been studied. In the United States, however, E. Rose (2004) observed that alongside the reversal of hypergamy, the disproportionate number of singles among the highest educated women also fell to more moderate levels (while Goldstein and Kenney (2001) even noted a reversal here too), but also that singlehood increased among the least educated men. Alongside the question of changing trends in hypergamy, it would be interesting to see whether the chances of finding a partner have followed a similar pattern in France.
II – Data from the 1999 Family History Survey
23Hypergamy can be measured via household surveys, but surveys providing information on first union formation across cohorts are all too rare. Here, we use data from the Family History Survey (Étude de l’histoire familiale (EHF), INSEE-INED, 1999 ; Lefèvre and Filhon, 2005), the most recent of the Family surveys to provide data on age at first union formation.  Conducted in conjunction with the population census, it was administered to a very large sample of 145,000 men and 235,000 women, so changes across cohorts can be examined in detail. Our analysis is limited to individuals born between 1920 and 1970, aged 29-79 at the time of the survey.
24The data have a major shortcoming however : as the partner’s educational level is obtained from the census individual form, it is only available for couples who were cohabiting at the time of the survey. This means that the first unions of individuals who have separated (and possibly repartnered) cannot be studied. This would introduce a slight bias if the couples who had separated were different from all first couples, for example, if they were mainly couples where the man was less educated than the woman, the traditional family model discouraging not only female hypogamy but also separation and divorce. This bias is especially problematic for measuring changes over time, since the proportion of individuals who have only lived with one partner varies across cohorts (Table 1), falling slightly for the cohorts born after the war (cohort effect). The proportion of individuals still living with their first partner follows the same pattern for men, but increases for women (age effect linked to higher mortality of their partners). The scale of this effect is relatively small, however, given the small number of couples concerned – especially if the oldest and youngest cohorts are excluded.
Distribution of individuals by conjugal history, sex and birth cohort*
Distribution of individuals by conjugal history, sex and birth cohort** “No missing values” means that the individual’s educational level and that of his/her first partner, the date of union formation and the individual’s date of birth are all provided. The following analyses never use all these items of information together : this column gives a theoretical minimum threshold for the samples used.
Interpretation : Among women in the sample born in 1920-1929, 36% are still with their first partner and can be described with no missing values for all the variables considered.
Coverage : Individuals born in France.
25In addition, there is a risk that the educational level at the time of the survey might not be a true reflection of the partners’ educational levels at the time of union formation, i.e. at the end of their initial education, as they may have obtained further qualifications since then. Such cases are quite rare in France, however. We also know that there is a reporting effect that varies with age (Baudelot, 1989), though it is limited in scale. We note that in her study of educational homogamy based on the same data, Mélanie Vanderschelden (2006) identified a downward trend consistent with the findings of our own study based on several cross-sectional surveys (Bouchet-Valat, 2014a) ; the bias introduced is not critical therefore. Nonetheless, to avoid the potential problem of over-interpreting minor changes over time, we will focus solely on the most clearcut trends. 
26We are interested here in the first cohabiting union that lasted for more than six months (definition used by the Family History survey), whether or not the couple were married ; the meaning of marriage changed so greatly over the study period that a study of first marriage would have had little value. While the meaning of cohabitation has also changed over cohorts – this type of union is no longer a simple prelude to marriage (Toulemon, 1997) – this definition seems the most appropriate for our study, since the aim here is to examine the first conjugal relationship. By limiting the study to first unions, we can model the process of union formation and its changes by cohort ; this would have been impossible if unions of different orders had been included.
27Our classification of educational levels is quite detailed, so that meaningful distinctions can be made within both the oldest and youngest cohorts. We nonetheless grouped the lower secondary diplomas (CAP, BEP, BEPC), since there is no clear hierarchy between these qualifications that remains valid for all cohorts. The categories are : no qualifications ; primary school certificate (CEP) ; lower secondary certificate ; general, technological or vocational baccalauréat ; undergraduate diploma (below bachelor’s level) ; bachelor’s or post-graduate diploma. An individual is considered to be endogamous, hypergamous or hypogamous if his/her educational level is, respectively, equal to, below or above that of his/her partner in the chosen classification. We do not take account of differences between educational programmes and subject choices, whose effect on occupational gender inequality is considerable (Duru-Bellat, 2004 ; Couppié and Epiphane, 2006), but which do not totally counteract the effect of vertical differentiation. Note that the qualification considered is the last one obtained by the individuals, whether before or after union formation.
28All the analyses use the weights supplied by the survey. The models are estimated in the R environment (R Core Team, 2014), in particular with an extension to the flexsurv package (Jackson, 2014). 
III – From female hypergamy to female hypogamy with regard to education
29In terms of education, the relative positions of men and women on the marriage market changed radically between the 1920 and 1960 birth cohorts. In response to this major structural change, were social norms able to withstand the formation of couples where the woman is more highly educated, or did couple composition simply adapt to the “stock” of available spouses ? This raises the question of the pertinence of social gender norms, and their capacity to withstand a rise in women’s status in a key area, that of social and cultural capital. Indeed, in societies where women generally spent less time in education than men, the prevalence of hypergamous couples could be explained without reference to individual preferences or the weight of social norms : structural constraints, themselves linked to social norms, were sufficient to produce this result – unless counteracted by singlehood.
30After describing raw changes in the educational composition of couples, we will examine whether they can be explained entirely by changing structural constraints, or whether behaviours have evolved at a deeper level.
Female hypogamy has become the most frequent situation
31We will begin by studying raw changes in the educational composition of couples (top of Figure 1). Note that the figures for men and women are not exactly identical, since a person can form a first union with someone who has already had a partner.
Relationship between educational levels of partners in a first union by year of birth (observed and expected proportions assuming random partner choice)
Relationship between educational levels of partners in a first union by year of birth (observed and expected proportions assuming random partner choice)Note : The observed percentages are smoothed using first-degree local regression, and the expected percentages are calculated on the basis of five-year cohorts. The figures for men and women are not exactly identical, since a person can form a first union with someone who has already had a partner.
Interpretation : In 36% of intact couples where the woman was born in 1940, the woman was less educated than the man, and in 23% the woman was more educated. These proportions would have been 44% and 33%, respectively, if the couples had formed randomly in the 1940-1944 five-year cohort.
Coverage : Intact first unions of individuals born in France.
32For both sexes, endogamous couples – those where both partners have the same level of education – were by far the most frequent (50% in our classification for the 1920 cohort). Couples where the man is more educated (female hypergamy or male hypogamy) were a minority, even among prewar cohorts (Guichard-Claudic et al., 2009).
33Gender asymmetry is only expressed in exogamous situations, via the difference between the proportions of couples where it is the man who is more educated (30% for the same cohort) and those where it is the woman (20%). Two phases of change across cohorts can be identified. In a first phase, corresponding to the cohorts born in the 1920s and 1930s, the proportion of endogamous couples declines in favour of both types of exogamous couples (hypergamous and hypogamous). This phase does not correspond to a clear change in gendered differences, although female hypergamy increases slightly faster than the reverse situation.
34In a second phase, beginning with the 1940-1945 cohorts, the proportion of couples where the woman is less educated starts to fall rapidly, while the proportion where she is more educated now increases at a steady pace. In the cohorts born in the late 1950s, cases of female hypogamy outnumber those of hypergamy. This trend leads to a situation where, among the cohorts born in the early 1970s, the proportion of first couples in which the women is more educated than the man is equal to that of endogamous couples (40%). And this will likely apply to the most recent cohorts likewise. The timing of these changes in hypergamy coincides perfectly with the many transformations that affected the family over the same period. In their history of the baby boomers, Catherine Bonvalet, Céline Clément and Jim Ogg (2015) show that these younger generations “plunged” into the breach opened up by the cohorts born during the Second World War. This image applies perfectly here.
35Changes in hypergamy among the medium-educated provide a striking illustration of the reversal that occurred around these pivotal generations. Among holders of a lower secondary qualification (CAP/BEP/BEPC) born in the 1920s, 27% of women, but just 7% of men, had a more educated first partner. These figures have been completely reversed, with 20% of women and 34% of men in the 1960s cohorts having a more qualified partner.
36The changes we have examined here are primarily the consequence of a rise in women’s educational attainment : since the cohorts born in the mid-1950s, women have been more educated, on average, than men, and the distribution of first unions that would be expected if individuals chose their partner at random among members of the five-year birth cohort of the opposite sex follows the same pattern (Figure 1, lower part).  The most notable difference between the observed and theoretical situation is linked to the fact that among observed couples in all cohorts, we find the classic situation where endogamous couples are overrepresented and the two other types of couple are underrepresented.
The reversal cannot be explained by structural changes alone
37To take things further, we control for the educational distribution of the male and female populations by means of log-linear modelling (Agresti, 2002). This technique allows us to measure the effects on union formation of social norms, individual preferences and opportunities for meeting potential partners, beyond the simple availability constraint of partners with a given educational level.
38The model uses as a reference a situation where individuals choose at random from among all potential partners in their five-year birth cohort (excluding singles considered to be outside this marriage market). The model reasons on the basis of union frequencies by both partners’ educational levels and the birth cohort of the reference individual (man or woman). It includes parameters controlling for marginal distributions by educational level of individuals of the sex considered and of their first partners for each birth cohort, and parameters measuring the over- or under-representation of hypergamous and hypogamous couples with respect to endogamous ones.
39For each sex separately, the model is founded on three variables. Each first couple is characterized by the man’s educational level M, that of the woman, W, and the birth cohort C of the reference individual, either a man or a woman depending on the population considered.
40The formula, applied separately to the first couples of men and women is written as follows :
42with mmwc being the numbers of couples corresponding to a given combination m, w, c of values for M, W and C. The first parameter λ is a constant. λMm (respectively λWw) controls for the proportion of couples where the men’s educational level M has the value m (the woman’s level W has a value of w).λCc controls for the proportion of couples where the reference individual is in cohort c. λMCmc and λWCwc measure the interaction between the variables M and W on the one hand, and C on the other, i.e. the proportion of individuals taken as reference in cohort c and of their first partners having an educational level m or w. The variables 1m>w and 1m<w are dummies corresponding, respectively, to situations of female hypergamy and hypogamy. λC(M>W)c and λC(M<W)c are coefficients of interaction between the variables M and W conditioned by cohort C, measuring the propensity of persons in each cohort to have a partner who is more or less educated than him/herself. These coefficients (shown on Figure 2) thus identify the over- or under-representation of these situations with respect to the endogamous configurations in the cohort : they correspond to the propensity (or otherwise) to hypergamy and hypogamy, with respect to endogamy.
Relative educational hypergamy and hypogamy among first unions by year of birth in a log-linear model
Relative educational hypergamy and hypogamy among first unions by year of birth in a log-linear modelCoverage : Intact first unions of individuals born in France.
43The estimate is made simultaneously for men and women, introducing interactions between the respondent’s sex and all the parameters described above. The overall model is based on 329,007 observations and has a deviance of 833,230 for 1,900 degrees of freedom.
44In Figure 2, the coefficients are associated with a quasi-confidence interval (Firth and Menezes, 2004), making it possible to compare all points with each other, and not only each point with the reference (0 for endogamous).
45The absolute trends observed in the previous section persist even after controlling for population distribution (relative hypergamy and hypogamy, Figure 2). The same phases can be distinguished, with a trend reversal in female hypergamy around the 1940-1945 cohort ; from the 1955-1959 cohort, the situations where the woman is more educated are as numerous as those where the man is more educated. These results show that the choice of partner follows and overtakes changes in the educational distribution of the population. In the cohorts born in the 1960s, couples in which the woman is less educated than her partner are less frequent than those in the reverse situation, by comparison with what would be expected if partners chose each other at random.
46In the case of France, our findings are consistent with the more general observation of A. Esteve, J. García-Román and I. Permanyer (2011) who found that absolute and relative hypergamy always point in the same direction. It appears that social norms, proxied here by the choices made by individuals beyond the constraints of partner availability, are merely the reflection of the most frequent couple configurations in a given cohort, observed in Figure 1. In other words, the growing proportion of couples where the women is more educated than her partner – an inevitable demographic consequence of the rise in women’s educational attainment – seems to have produced a change in the preferences of individuals, in the social norms that are internalized or imposed upon them by their social environment. It is surprising to observe the relative ease with which gender norms seem to have given way, and even changed direction : it would appear that male dominance is based above all on structural factors, with normative constructions being superimposed upon a de facto situation.
47If it is as deep-seated as it appears, this change in hypergamous behaviour must have been accompanied by an equally remarkable transformation in the chances of finding a partner for the most disadvantaged individuals on the marriage market, namely low educated men and highly educated women. In fact, this is only partially the case.
IV – Singlehood less prevalent among highly educated women, but more so among low educated men
Modelling first union formation across cohorts
48The study of permanent singlehood, i.e. the fact of never having formed a union, is quite problematic, since the phenomenon cannot be perfectly observed until the deaths of the individuals in question. Changes over time can therefore only be analysed with a very long time lag. This difficulty can be limited by measuring the proportion of persons ever in a union at a given age, but changes in the age at first union formation cannot be taken into account : a delay in union formation increases the proportion of singles observed at a given age, although this increase may be offset by a greater propensity to form unions at older ages.
49To measure permanent singlehood as distinct from this age effect, we use an accelerated failure time log-logistic model, one of the most effective techniques for predicting, with a relatively high level of confidence, the proportion of individuals who will never form a union in their lifetime, based on the observation of a cohort up to a given age. As first union formation slows down considerably beyond age 35, a reliable estimate of the proportion of permanent singles can be obtained for the cohorts born in the 1960s, and trends for the more recent cohorts can also be outlined.
50The model used here (Brüderl and Diekmann, 1995) includes three parameters, estimated separately for each educational attainment group : the proportion permanently single, the median age of union formation, the shape of the union formation age distribution. Within each educational group, these parameters are allowed to vary across cohorts by means of restricted cubic splines with three degrees of freedom (Keele, 2008) which provide a flexible way to study changes over time. We know that age at union formation has not evolved in a linear manner, but has followed a U-shaped curve ; as found here, the baby-boom cohorts had the lowest median age at first union formation in the twentieth century (Prioux, 2003). The very good match between the log-logistic distribution and that of first unions gives an indication of the quality of the predicted proportion of permanent singles ; this distribution, very similar to the Coale-McNeil distribution classically used in studies of marriage, has been applied for this purpose by various scholars (Keeley, 1979 ; Brüderl and Diekmann, 1995, 1997). The extrapolated proportion of permanent singles in the most recent cohorts obtained here is based first, on the behaviour of these cohorts observed up to the survey date, and second, on its extension using the splines of observed distributions for the older cohorts.
Changes in the proportion of permanent singles by educational level
51These results confirm the existence of a strongly gendered model of union formation in the prewar cohorts. Among men, only those with no qualifications or a primary certificate are less likely to form a union than other groups (difference of 5 and 10 percentage points, respectively, in the 1940 cohort).  No difference is observed between the other educational levels (Figure 3). Among women, on the other hand, a very clear gradient is observed ; the likelihood of forming a union decreases steadily as their educational level increases (falling from 95% to 75%). This corresponds to a situation where hypergamy is the norm.
Proportions of individuals who will have experienced at least one union in their lifetime, by year of birth and educational attainment
Proportions of individuals who will have experienced at least one union in their lifetime, by year of birth and educational attainmentInterpretation : 75% of women born in 1920 and with an undergraduate diploma have lived in a union in their lifetime.
Coverage : First unions of individuals born in France.
52But here too, this model of union formation no longer prevails among the most recent cohorts, and there are very noticeable changes across cohorts in the differences between educational groups. Male permanent singlehood increases among the lowest educated from the 1920 cohort, then in all groups from the 1940 cohort, but less so among the most educated men. Among women, the differences between educational groups grow progressively narrower, before disappearing altogether. From the 1950 cohort onward, the most educated women are just as likely to form a union as the least educated, with the exception of women with a bachelor’s degree or higher, who catch up slightly later.
53Like Maria Winkler-Dworak and Laurent Toulemon (2007), we observe that for the least educated women, the likelihood of forming a union decreased with respect to those who are more highly educated. But the main change for women appears to be the increasing likelihood of union formation among the highly educated.  We also confirm their observation that gender differences have disappeared for all educational levels, with the exception of the lowest educated men, whose chances of forming a union remain lower. Last, these findings on permanent singlehood are consistent with measures of singlehood at a given moment, in terms of both educational level and occupational category. In 2006, among both men and women in their thirties, only the lowest educated had a lower rate of union formation (Daguet and Niel, 2010), while in the 1990s this rate decreased with educational level (Albouy and Breuil-Genier, 2012). Likewise, in 2008, women in their thirties in higher-level occupations more often had a partner than female manual and clerical workers, while the reverse was true in 1990 (Buisson and Daguet, 2012).
54These observations tie in directly with the changes commented in the previous section. In both absolute and relative terms, the decline in female hypergamy began with the cohorts born around 1945. It was these same cohorts that, for women, saw the almost total disappearance of differences in permanent singlehood linked to educational level and that, for men, initiated the rise in permanent singlehood for all educational groups. The fundamental link between female hypergamy and singlehood among the most highly educated women is thus confirmed : educated women’s disadvantage on the marriage market disappeared at the same time as the imbalance created by the preference for female hypergamy.
55Educational expansion may have played a major role in this development. As few women in the prewar cohorts held a degree in higher education, the high levels of permanent singlehood in this group affected only small numbers of people, so the ricochet effect on men was limited in extent, with only a small proportion being unable to find a partner. By contrast, among the most recent cohorts, the persistence of such high levels of singlehood among highly educated women (around 20-25%) was no longer sustainable, given the large numbers involved : the effect would have been a severe marriage market imbalance, with educated men forced to marry women from social groups very distant from their own, and all low educated men condemned to a life of singlehood. In addition, women with degrees in higher education born before the war cannot be directly compared with their younger counterparts born after 1950 : they formed a highly selected population which was often forced to choose between a career or a family. With the expansion of female education, this dilemma is less of an issue for most women.
56With the increase in singlehood among low educated men, the convergence between men and women is not complete. While the chances of finding a partner no longer vary by educational level among the most recent female cohorts, we note the emergence and development of an educational hierarchy among men : the chances of forming a union now increase with educational level. This scale is the exact reverse of that which prevailed among women born before the war. Far from indicating a gender reversal, as observed previously with respect to hypergamy, it reveals the persistence of a gendered model in which the man’s social status is more important than that of his partner. Indeed, with the expansion of education, qualifications have becoming an increasingly important factor on the job market. As a consequence, the lowest educated – all the more so given that their numbers have shrunk – carry the greatest burden of risk in terms of poor labour market outcomes and long-term employment insecurity (Gasquet, 2003). In particular, these difficulties affect men’s union formation by delaying their departure from the parental home, while women, on the contrary, tend to fall back on their conjugal or family life (Galland, 1995 ; Jaspard et al., 1995).
57This article has shown that between the cohorts born before the Second World War and those born in the 1970s, female educational hypergamy in first cohabiting unions first declined and was later reversed. This trend affected the three dimensions of hypergamy : absolute hypergamy, relative hypergamy and the gendered distribution of singlehood by level of education. First, following the rise in women’s educational level, which now exceeds that of men, the woman is more often the most highly educated partner in first unions between persons born since the late 1950s. Second, even after controlling for changes in the educational distribution of the population, a similar reversal of relative hypergamy is observed. This finding indicates that beyond the constraints of availability of partners with a given educational level, individuals in recent cohorts “choose” (more frequently than would occur if partners were randomly selected), to form a union in which the woman is more highly educated than the man, rather than the reverse. Third, rates of female permanent singlehood, which increased markedly with educational level among the prewar cohorts, have become very similar for women of all educational levels since the cohorts born in the 1960s. These three trends indicate that the shift in gender norms is much stronger than a mere mechanical adjustment in response to the expansion of female education.
58Like the increase in women’s educational level, the observed change appears to be both a sign and a factor of the transformation of gendered roles within the couple. This phenomenon is creating the potential foundations for a long-term shift in attitudes via the socialization of new generations within families where the woman is the most highly educated partner.
59Even more remarkably, the speed of the observed changes suggests that these norms have been swept away without the slightest resistance. Comparing successive cohorts in a single country, this study confirms the results obtained by comparing different countries at different dates, namely that in no society is a relative preference for female hypergamy observed that goes against the educational distribution of the male and female populations (Esteve et al., 2011). This result tends to cast doubt upon the very existence of a norm of female hypergamy, with regard to education at least. More generally, it raises the question of whether a cultural system is capable of withstanding a profound transformation of the structural conditions underpinning its creation : can a norm continue to be meaningful and relevant if it is no longer anchored in practical reality ? Here we find the dual meaning of the word “normal” : the positive meaning of that which is most frequent, and the normative meaning of that which should be, with the former appearing to determine the latter ; in the terms of Pierre Bourdieu (2001), “the spontaneous statistic through which each of us forms our representation of what is normal”.
60Yet the spectacular change observed here must be placed in perspective. First union formation is still highly gendered, even within the most recent cohorts. It is still more difficult for low educated men to find a partner, indicating a persistent negative effect of labour market difficulties on male union formation. Moreover, and more fundamentally, the consequences of the development of female educational hypogamy are not clear (Guichard-Claudic et al., 2009). We know that despite women’s rising level of education, their careers are still not equivalent to those of men. This is the consequence of persistent gendered segregation by skill type at equal levels of qualification (Caille et al., 2002 ; Duru-Bellat, 2004), but also of segregation on the labour market (Couppié and Epiphane, 2006).
61Neither has this change had any profound effect on the division of domestic labour. In itself, a woman’s status as a student or graduate of higher education is not sufficient to establish a position of superiority within the couple or to renegotiate gender roles (Lefeuvre, 2008) ; only the woman’s occupational status appears to have a certain effect in this regard (Bauer, 2007 ; Champagne et al., 2015), at least when the woman has a higher income than her partner (Nicole-Drancourt, 1989 ; Ponthieux and Schreiber, 2006). Women’s careers are often held back by the fear, shared by both partners, that the woman will earn more than the man (Testenoire, 2008). The gender division in employment seems to be more fundamental in the definition of gendered identities than the partners’ relative levels of cultural capital. This would explain why gendered roles persist despite the reversal of educational hypergamy – which remains a mirage so long as its effects are not translated into “hard cash”. But there is still plenty of potential for change : female educational hypogamy increases the likelihood of the woman being the couple’s main breadwinner (Klesment and Van Bavel, 2015), and here again, a slow convergence may well be taking place (Morin, 2014).
French Institute for Demographic Studies (INED), Center for Studies in Social Change (OSCSciences Po & CNRS), Quantitative Sociology Laboratory (LSQ-CREST), France.
Correspondence : Institut national d’études démographiques, 133 boulevard Davout, 75020 Paris, email : firstname.lastname@example.org
Like most household surveys, Family and Housing 2011 (Famille et logements 2011) only includes information on current unions : it cannot be used effectively to study the process of first union formation and permanent singlehood.
Given the limitations of the Family History survey, we decided to use data from a second survey, Life Event Histories and Entourage (Biographies et entourage, INED, 2001 ; Bonvalet and Lelièvre, 2016), which provides detailed information on the conjugal trajectory of respondents and their successive partners. However, this smaller survey (2,830 respondents) is not always able to provide statistically significant results ; nonetheless, when they are significant, the results obtained confirm those presented here. The analyses based on the Life Event Histories and Entourage survey can be found in a working document (Bouchet-Valat, 2014b).
The code for estimating the models used here is available online on the author’s personal webpage, and can be supplied on request.
The theoretical situation of random union formation used here assumes that all individuals form a union within their five-year birth cohort, and excludes the option of singlehood. It thus only represents a reference accounting for the influence of the population structure.
The apparent convergence before 1930 between the curve of the least educated men and the other curves appears to be attributable to a selection effect arising from differences in mortality by social background beyond age 70.
This trend was also observed by Isabelle Robert-Bobée and Magali Mazuy (2005) at age 45 for cohorts born before 1955.