Different theoretical perspectives can explain why a continuously negative relation between educational level and fertility would be observable in most low-fertility contexts. First, longer enrolment in education leads to the postponement of births (Ní Bhrolcháin and Beaujouan, 2012). This is connected to the notion that postponed births lead to lower completed fertility among women because they are nearing the so-called social age deadlines (finding themselves too old to have children, given existing norms) and biological deadlines (not being able to have children) (Schmidt et al., 2012).

Second, a diffusion of more individualistic norms and gender equality through higher education has been cited as a reason why women’s agency has increased while their fertility has declined (Basu, 2002). From a cultural perspective, educational attainment could be connected to less traditional attitudes and weaker interest in forming a family, as the rise in education is associated with the diffusion of individualisation and secularism (Lesthaeghe, 1983, 2014). Higher education and less traditional values are associated with more positive attitudes towards voluntary childlessness across European countries (Merz and Liefbroer, 2012), and perceived costs to individual autonomy are negatively related to parenthood (Liefbroer, 2005).

Third, following economic theories of the family, the investment in education leads to higher returns in terms of income and status. Perceived career costs have been shown to influence the timing of women’s entry into parenthood (Liefbroer, 2005). Connected to this, another important emerging factor could be the growing number of women with higher education and the resulting lower number of desirable partners with similar or higher levels of education (Van Bavel, 2012). As women are now more likely to be the main breadwinner of their family than in the past (Klesment and Van Bavel, 2017), the opportunity costs of childbearing may have increased. Furthermore, given that highly educated women tend to prefer partners with a similar education level, and that there is no longer a sufficient supply of equally educated potential male partners – also referred to as the “mating squeeze” (Van Bavel, 2012) – it could be expected that women with advanced degrees are becoming more likely to remain single and childless (Van Bavel, 2012).

In summary, several mechanisms might explain a negative educational gradient in fertility. This is evidenced by previous empirical studies. For cohorts born between 1940 and 1961, Wood et al. (2014) showed that childlessness is most frequent among highly educated women in 14 low-fertility countries. A study of 25 European countries found a persistent, although gradually decreasing, negative educational gradient in completed fertility (Merz and Liefbroer, 2017).

Recent evidence, however, indicates that fertility is declining, especially among lower educated individuals. A positive educational gradient has been observed for men, with highly educated men having more children than less educated men, e.g. in Finland (Nisén et al., 2014), and lower educated men having a lower probability of selecting into partnerships, and consequently entering parenthood, across 10 European countries (Trimarchi and Van Bavel, 2017). While this has been a longstanding trend for men, higher levels of childlessness are also increasingly found among lower educated women, primarily in the Nordic region (Jalovaara et al., 2019).

As empirical evidence shows, being in a partnership continues to be considered a prerequisite for having children (Berrington, 2004; Harknett and Hartnett, 2014; Kapitány et al., 2012; Rybińska and Morgan, 2019). Single men and women are significantly more likely to postpone childbirth and abandon childbearing plans in comparison to married and cohabiting respondents in the Netherlands, Switzerland, Hungary and Bulgaria (Kapitány et al., 2012). Furthermore, people who have never had a partner or are in several living apart together relationships during their life course are most likely to remain childless in Germany (Raab and Struffolino, 2020) and Finland (Jalovaara and Fasang, 2017; Saarela and Skirbekk, 2020).

There are reasons to expect that women’s higher educational attainment and higher earning potential are increasingly becoming key factors in the partner choice process of men. In younger generations, women’s enrolment in education now exceeds that of men in many countries (Van Bavel et al., 2018; Vincent-Lancrin, 2008). While educational expansion has led to higher overall levels of homogamy (both partners have higher education), the gender imbalance in higher education has contributed to a rise in hypogamy, i.e. the woman is more educated than the man (Katrňák and Manea, 2020; Leesch and Skopek, 2024). In other words, if the partners in a couple have differing levels of educational attainment, the woman is now more likely to have a higher degree than the man (De Hauw et al., 2017; Esteve et al., 2016; Grow and Van Bavel, 2015). Across the 1954–1980 cohorts and 27 European countries, Erát (2021) indeed found a positive correlation between the share of women in higher education and an increase in hypogamy. As the share of women with higher education increases, higher educated women are more likely to “partner down” while lower educated men are more likely to partner with higher educated women, possibly leading to the increased marginalisation of lower educated women in the partnering market.

Changes in the partnering market and preferences among men for having an equally or higher educated female partner can also be explained by women’s increased earning potential and the diffusion of gender egalitarian norms. The labour market position of women compared to that of men has improved in recent years (Goldin, 2014), and the reversal of the gender gap in education is associated with women having higher earnings relative to men (Van Bavel and Klesment, 2017). Furthermore, the stability and security of typically male, blue-collar occupations have declined, leading to a rise in dual earner and female breadwinner households (Klesment and Van Bavel, 2017; Vitali and Arpino, 2016). Along with the increasing labour force participation of women, we see a diffusion of gender egalitarian norms, which are, in turn, associated with changing gender relationships in both the public and the private sphere (Goldschneider et al., 2015). Men holding more egalitarian gender role attitudes are, for example, more prone to sharing household chores (Sullivan et al., 2014) and more likely to enter a union with a higher educated female partner (Trimarchi, 2022). Measuring actual mate preferences over 25 years in the US, Boxer et al. (2015) found that both men and women place increasing value on a partner’s financial prospects. While the traditional preference for a financially successful male partner might persist, research also shows that women increasingly value family orientation and communion in their male partners (Meeussen et al., 2019), indicating increasing gender convergence in partner preferences (Zentner and Eagly, 2015).

Finally, the opportunity costs of childrearing have decreased as women are given more opportunities to combine family and work life (Kalwij, 2010). This could benefit women with higher incomes in particular, as they are, for example, more likely to take advantage of (higher quality) childcare (Van Lancker, 2018). Studies indeed suggest that some work–life reconciliation policies, such as those providing for parental leave and childcare services, are especially beneficial to women with higher, rather than lower, incomes (Bergsvik et al., 2021; Wood et al., 2020).

These developments imply that the earning potential of women is an increasingly important factor in the partnering market. In this study, we are interested in the overall likelihood of living with a partner versus remaining single until a certain age. Rather than analysing differences between marriage and cohabitation, which also vary by region (Hiekel et al., 2014), we therefore jointly consider both married and unmarried cohabitation. While differences in the educational gradients of marriage and cohabitation exist (Kalmijn, 2013; Perelli-Harris and Lyons-Amos, 2016), findings for more recent cohorts show an overall lower prevalence of unions among lower educated women, especially in more gender equal contexts (Hudde and Engelhardt, 2023; Jalovaara, 2012; Jalovaara and Andersson, 2023; Sandström and Karlsson, 2019; Sturm and Van Bavel, 2024).

In Finland, higher education has been shown to promote both marriage and cohabitation among women born between 1969 and 1981, with an even stronger positive effect of higher education and income being observed in the most recent cohort born between 1975 and 1981 (Jalovaara, 2012). From 1989 to 2020, the prevalence of unions between ages 18 and 49 declined across all educational groups in Finland, but especially for lower educated men and women (Jalovaara and Andersson, 2023). In line with these findings, lower educated women have been shown to be less likely to live with a partner in Sweden, the Netherlands and Estonia in the years 2002 to 2013 (Sandström and Karlsson, 2019). Increasing rates of living without a partner have also been found for lower educated women in Western Germany over the 1976–2019 period (Hudde and Engelhardt, 2023). Across the 1960–1985 cohorts, the likelihood of ever having cohabited (both married and unmarried) with a partner by ages 35 to 45 has decreased for lower educated women compared to higher educated women in the southern and western European regions (Sturm and Van Bavel, 2024). As partnerships continue to be a requirement for having children, a lower likelihood of having a (stable) partnership by a certain age may, in turn, make it more likely that women will postpone or forego parenthood.

In summary, we argue that the educational gradient in fertility may be shifting, as higher education and, consequently, higher earning potential improve women’s chances of finding a stable partner in recent cohorts. Our argumentation is based on shifts in the share of women with higher education, assortative mating and changing preferences of men. Nevertheless, the link between education, fertility and couple formation remains complex. Unobserved factors such as certain personality traits and family orientation may influence the likelihood of both entering a stable partnership and having children (Baizán et al., 2003; Trimarchi and Van Bavel, 2017). In a context in which growing numbers of women are attaining higher education, lower educated women represent an increasingly selective group. Having children earlier might limit their opportunities to pursue higher education (Johansen et al., 2020; Raymo et al., 2015), leading to a negative association between education and fertility. In contrast, individuals who prefer to have a higher educated partner or a higher income before starting a family could also increase their efforts to achieve higher education, leading to a positive association between education and fertility. In conclusion, the advantages of higher education in the partnering market may explain why fertility is declining among women with lower levels of education. The selection into lower education, on the other hand, could explain both a positive and a negative gradient in fertility.

The share of women in higher education differs across European countries and regions, as do women’s labour market participation and income levels. Furthermore, family policies, and the degree to which they promote gender egalitarianism or reinforce the male breadwinner model, differ across contexts. Consequently, we can expect to observe differences between countries and regions in how education relates to fertility, and how these patterns shift across cohorts.

The Nordic countries in our sample (Denmark, Finland, Iceland, Norway and Sweden) are characterised by relatively high shares of women in tertiary education and high levels of female labour force participation (Eurostat, 2023a, 2023b). The investments in these countries in both care provided to children under the age of three and parental leave are among the highest in Europe, making it easier for women to combine work and family life (Thévenon, 2011). The clearest signs of a converging or reversing educational gradient in fertility have been found in the relatively gender egalitarian context of the Nordic countries. Jalovaara et al. (2019) compared cohort total fertility and ultimate childlessness by educational attainment for the cohorts born between 1940 and 1974. For women in Denmark, Norway and Sweden, they found that childlessness levels are now highest among those who are the least educated (ISCED 0–2). Higher childlessness levels and stronger declines in first births have also been found for lower educated women in more recent cohorts in Finland (Hellstrand et al., 2022; Rotkirch and Miettinen, 2017).

Moreover, recent studies have reported an emerging positive association between individual economic stability and fertility in the Nordic countries. For example, Comolli et al. (2021) observed a larger decline in first and second birth rates among low educated women after the economic crisis in Denmark, Finland, Norway and Sweden. Stronger declines in first birth rates have also been found among women with weaker labour market attachment in Sweden (Ohlsson-Wijk and Andersson, 2022), and among women with education in fields characterised by higher unemployment and lower income levels in Finland (Hellstrand et al., 2024).

Compared to the Nordic countries, the western European countries in our sample (Austria, Belgium, Switzerland, Germany, France, Great Britain, Ireland and the Netherlands) have family policies that are more strongly influenced by the one earner family model, although considerable variation exists among them (Thévenon, 2011). While women’s levels of participation in higher education and the labour force are, on average, lower in these countries than in the Nordic countries, they have increased (Eurostat, 2023a, 2023b), and levels of gender equity in financial and economic resources are higher than the EU average across all the countries in our sample (EIGE, 2023).

The heterogeneity of the western European context is reflected in the educational gradient in fertility. In the German-speaking countries (Austria, Germany and Switzerland), where childlessness rates are among the highest in Europe, higher educated women are more likely than their lower educated counterparts to remain childless (Neyer et al., 2017; Sobotka, 2012). However, for Germany, Kreyenfeld and Konietzka (2017) found that childlessness is now increasing more among lower educated women born in West Germany between 1940 and 1964, while it has plateaued for women with a tertiary degree.

Signs of a diminishing educational gradient in fertility are also found in France, Belgium and the Netherlands. For example, Compans (2021) showed that among the cohorts born between 1950 and 1970 in France, higher educated women tend to catch up on fertility, while lower educated women are more likely to remain childless. For Belgium, empirical evidence points to a weakening, albeit still positive, educational gradient in childlessness (Van Bavel, 2014; Wood et al., 2014). For the Netherlands, van Wijk, Valk and Liefbroer (2022) have shown that for women, employment precariousness and past income precariousness are negatively related to the likelihood of transitioning to parenthood. Conversely, women’s incomes have been found to be increasingly positively related to higher first birth transition rates over the 2008–2022 period (van Wijk 2024). For the United Kingdom, the findings are mixed. Kuang et al. (2024) identified a persistent positive educational gradient in childlessness among women born between 1956 and 1978, while Ermisch (2021) reported a more pronounced decline in first birth rates among lower educated women born after 1970. Based on the increasingly gender egalitarian context of the western European countries and results pointing to changes in the negative gradient in fertility in some countries, we would expect to find an overall weakening of the educational gradient in fertility in these countries, albeit to a lesser extent than in the northern European countries.

The southern European countries in our sample (Croatia, Cyprus, Italy, Portugal, Slovenia and Spain) are generally characterised by welfare systems that rely more strongly on the family and provide less extensive childcare and more limited periods of paid parental leave (Merz and Liefbroer, 2017; Thévenon, 2011). Compared to the Nordic and western European countries, the shares of women enrolled in tertiary education and female employment rates are lower in southern Europe (Eurostat, 2023a, 2023b). Furthermore, in these countries, levels of gender equity in access to financial and economic resources are, on average, lower than the EU mean (EIGE, 2023). This could indicate that, compared to the Nordic and western European regions, the southern European region has a more traditional division of labour, less favourable conditions for combining work and family life and, consequently, higher opportunity costs when starting a family for women.

Previous studies indeed found positive gradients of childlessness for Spain and Italy (Donno and Tanturri, 2020; Impicciatore and Zuanna, 2017; Wood et al., 2014), as well as strong negative educational gradients in completed fertility for Cyprus, Spain and Portugal (Merz and Liefbroer, 2017). Nevertheless, for Spanish women born between 1965 and 1969, Reher and Requena (2019) found that differences in childlessness levels between female college graduates and lower and medium educated women disappeared or were considerably reduced. Based on previous empirical findings and the more traditional gender role orientation in the southern European countries, we would expect to observe that the educational gradient in fertility remains negative, although signs of a convergence across educational levels might be visible in more recent cohorts.

The eastern European countries in our sample (Bulgaria, Czechia, Estonia, Hungary, Lithuania, Latvia, Poland, Slovakia and Ukraine) are characterised by partly contradictory developments. On the one hand, higher education and female employment expanded rapidly during the communist period and, by 1975, there were no differences in terms of years of schooling between men and women in Bulgaria, the Czech Republic, Hungary, Poland and Slovakia (Ganzeboom and Nieuwbeerta, 1999). At the same time, fertility was promoted through strong pronatalist policies (David et al., 1999). Current employment levels across the eastern European countries are, on average, higher than those in the southern and western European countries and are comparable to those in the Nordic countries (Eurostat, 2023a). While women’s employment levels are high, levels of gender equity in women’s access to financial and economic resources in eastern Europe are considerably lower than the EU average (EIGE, 2023). Declining childcare provision following the fall of socialism indicates a re-traditionalisation of public policies (Pascall and Manning, 2000; Robila, 2012), and implies that women in eastern Europe face greater obstacles to combining childcare and work.

Compared to other European regions, childlessness levels in eastern Europe are relatively low (Beaujouan et al., 2016). Within the region, childlessness levels are highest among highly educated women born between 1916 and 1965 in the Czech Republic, Hungary, Poland, Romania and Slovakia, and the childlessness gap between lower and higher educated women even seems to have increased in more recent cohorts (Beaujouan et al., 2016). According to Wood et al. (2014), the educational gradient in first births is persistently negative for Hungary but is weakening for Bulgaria and Estonia. Overall, we would therefore expect to observe a persistent negative, albeit weakening, educational gradient in fertility across the eastern European region.

As empirical results show, there are signs that the negative educational gradient in fertility is changing in some countries. Nevertheless, results differ across countries, which is partly attributable to the different data sources and cohorts studied. By utilising data from the ESS, with round 9 of the ESS collected in 2018/19, we can include information on the age at first birth of more recent cohorts and compare the likelihood of transitioning to parenthood across broader geographical regions.

The current study aims to investigate how the likelihood of becoming a parent by a certain age cut-off varies by educational background, and whether this relationship varies across cohorts. Based on the theoretical framework and previous empirical findings, we first hypothesise that across cohorts, the likelihood of having a first child by age 35 (40) has decreased for lower educated women compared to medium and higher educated women ( Hypothesis 1 ). Second, we expect that changes in the association between educational attainment and fertility are influenced by the selection into partnerships. As having higher education is increasingly beneficial to women in the partnering market, we expect that women with lower education are less likely to find a partner to cohabit with, which, in turn, decreases their likelihood of entering parenthood. We therefore hypothesise that the association between educational attainment and having a first child is partly mediated by the likelihood of ever having cohabited with a partner (married or unmarried cohabitation) ( Hypothesis 2 ). Finally, based on the theoretical framework and previous findings, we expect that shifts to a positive educational gradient in fertility are most pronounced in regions with higher shares of women with tertiary education and higher levels of gender equality in terms of family policies, labour market participation and earnings. Especially in those regions, we would expect that having higher education as a woman is more conducive to finding a partner and entering parenthood. We therefore hypothesise that shifts to a positive educational gradient in fertility are especially pronounced in the Nordic region and, to a lesser extent, in the western European region, and are less visible in the southern and eastern European regions ( Hypothesis 3 ).

We use retrospective microdata from the ESS rounds 3 and 9, with fieldwork conducted between 2006 and 2007 for ESS round 3 and between 2018 and 2020 for ESS round 9. Both rounds include the “Timing of Life” module, which asks respondents about the timing of events such as first cohabitation and age at first birth. To test the likelihood of having had a partner and a first birth by age 35, we select women born between 1955 and 1985 to ensure they are at least 35 years old at the time of the interview. Additionally, to test to what extent the results differ for women who are at least age 40, we select a second sample that includes women born between 1950 and 1980 to ensure that they are at least 40 years old at the time of the survey. This means we can include first births happening between ages 35 and 40, which are rarer but increasingly common in recent cohorts. In the following, we focus on the sample, descriptive and multivariate results for the sample of women born between 1955 and 1985, who are at least 35 years old at the time of the survey. Results on the comparison between a first birth by age 35 and a first birth by age 40 are reported at the end of the results section.

In the ESS, we only have information on the year in which an event happened, rather than the month. For cohabitation, we limit the cut-off age to 34, counting births from those cohabitations until age 35, allowing for a pregnancy period. We also experimented with different age cut-offs, which led to similar results. We exclude women who report currently being enrolled in education (2.5% of the sample).

The ESS provides broad geographical coverage and the data quality is considered to be high due to standardised sampling strategies, non-response handling and data cleaning across countries (Stoop et al., 2010). We focus our analysis on EU countries but also include Great Britain, Iceland, Norway, Switzerland and Ukraine. In total, our sample includes women from 28 European countries. Not all countries participate in both rounds and the sample of lower educated women is increasingly limited as we move to more recent cohorts (for country participation by round, see Table S.1 in the supplementary material, available online at https://doi.org/10.1553/p-kddn-zj78). This limited number of observations is the reason why we group countries into four broader geographical regions to get robust estimates. Nevertheless, we include country dummies in our models and cluster standard errors by country to account for country heterogeneity within regions. Table 1 gives an overview of the sample size by country.

We run logistic regressions separately for countries grouped by geographical region. We control for country and estimate clustered standard errors by country to account for differences within regions. For the first outcome, cohabiting by age 34, we estimate two models. Model 1 includes all the main independent and control variables. Model 2 adds an interaction between education and cohort. For the second outcome, having a first birth by age 35, we estimate three models. Similar to the first outcome, we run the analysis with all the main dependent and control variables (Model 1) and then include an interaction between education and cohort to test the first hypothesis (Model 2). Finally, we run the analysis including the measure of having cohabited with a partner by age 34 to test the second hypothesis (Model 3). By comparing the coefficients before and after controlling for cohabitation, we can assess to what extent the association between educational attainment and fertility changes when accounting for the selection into cohabitation. We present average marginal effects (AMEs), as they have the advantage of being comparable across models. Unlike odds ratios, AMEs are not affected by changes in the distribution of independent variables (Mood, 2010).

The causal relationship between education, partnerships and fertility is complex and is affected by endogeneity. Unobserved factors such as certain personality traits could increase the likelihood of completing higher education as well as of finding a partner or entering parenthood. Not accounting for those factors could lead to biased estimates. To control for unobserved factors influencing both outcomes, previous studies have applied multi-process modelling (Baizán et al., 2003; Trimarchi and Van Bavel, 2017). Given that the ESS does not have information on all partnerships and fertility histories, it is not possible to apply a multi-process event history model. As a robustness check, to account for endogeneity issues between cohabitation and parenthood, we applied a bivariate probit model using the share of equally educated male partners as an instrumental variable. The instrumental variable should directly affect the likelihood of cohabiting while only influencing the outcome of a first birth by age 35 indirectly, through the effect on union formation. To run our model, we use the ivprobit command in STATA. The results are in line with those we obtained with a logistic regression analysis and show an overall negative influence of education on first births (see Section S.1 in the online supplementary material for the results of the IV approach and a more detailed explanation of the instrumental variable). The results of the IV probit give only marginal evidence of endogeneity, which could be due to the instrument used. It is difficult to argue that the share of male partners with the same level of education affects cohabitation directly, and first births only indirectly (through partnership formation). Educational homogamy may directly affect the transition to a first birth, given that similarly educated partners may share values, norms and preferences regarding fertility (Nitsche, 2024; Trimarchi and Van Bavel, 2018).

Figure 1 shows the share of women who experience a first birth by age 35 by region and educational level, and for the three cohorts in our sample. The majority of women experience a first birth before the age of 35 (81% across the sample), but the shares vary by region, cohort and educational level. In the Nordic region, approximately 82% of women experience a first birth before age 35. While lower educated women have the highest shares of first births by age 35 in the 1965–1974 cohort (91%), the shares converge across educational groups in the most recent cohort. In the western European region, on average, 77% of women experience a first birth by age 35. In the most recent cohort, the share of women having a first child by age 35 is higher for those with lower than with higher education (83% vs 74%). In the southern European region, only 64% of highly educated women transition to parenthood before age 35 in the most recent cohort. Among lower educated women in this cohort, the share is approximately nine percentage points higher (73%) but has declined 15 percentage points from the oldest cohort. Finally, the eastern European region has the highest share of first births by age 35 (∼88%). For highly and medium educated women in this region, the share having a first birth by age 35 has decreased by around 11 percentage points, while for their lower educated counterparts, it has increased by around six percentage points.

Figure 2 shows the share of women who have experienced a first cohabitation by age 34. The majority of women have cohabited at least once before the age of 34, but we see signs of a reversal in the educational gradient. For the western European region, we find that the share of women cohabiting has declined from 87 to 80% for lower educated women in the most recent cohort, while it has remained more stable or even increased for medium and higher educated women. For the southern European region, we also find a slight increase in the share cohabiting by age 34 among higher educated women (from 81 to 85%), while we see a decline among lower educated women (11 percentage points, from 89 to 78%). For the Nordic and eastern European regions, we observe more similar shares of women cohabiting by age 34 across educational levels and cohorts.