In recent decades, educational expansion has been a main driver of fertility postponement in countries across Europe (Neels and De Wachter, 2010; Neels et al., 2017; Ní Bhrolcháin and Beaujouan, 2012). At the same time, educational expansion has proceeded unevenly across population subgroups, particularly among the growing and heterogeneous group of individuals with a migration background. Hence, the impact of education on future trends in the timing and quantum of order-specific fertility will increasingly be shaped by the education-fertility nexus among individuals with a migration background, who are often still characterised by lower educational attainment and patterns of early family formation. To date, the education-fertility link on the one hand, and the fertility of the descendants of migrants on the other, have been studied largely in isolation from each other. However, understanding their intersection is becoming increasingly important.

The differentiating role of education in both the tempo and quantum of fertility in (western) European countries has been well established in previous research (Lappegard and Ronsen, 2005; Neels and De Wachter, 2010; Ní Bhrolcháin and Beaujouan, 2012; Vasireddy et al., 2023). Higher educational attainment has consistently been associated with postponement of the transition to parenthood and higher levels of childlessness (Gustafsson et al., 2002; Martin, 2000; Mills et al., 2011; Rendall et al., 2005; Wood et al., 2014), although recent studies report decreasing educational gradients in childlessness among more recent cohorts (Jalovaara et al., 2019; Köppen et al., 2017; Rybinska, 2020). After entry into parenthood, higher educated women have often been found to transition more rapidly – and sometimes more frequently – to second and higher order births than their lower educated peers. This pattern has been repeatedly described as reflecting selective entry into parenthood, whereby only the most fertility-motivated among the highly educated individuals become parents in the first place (Klesment et al., 2014; Köppen, 2006; Kravdal, 2001; Neels and De Wachter, 2010; Wood et al., 2020, 2025). Largely separate from research on the association between education and fertility, a growing literature has documented substantial variation in fertility patterns across origin groups and migrant generations. While first generation migrants generally exhibit higher fertility than natives (Guarin Rojas et al., 2018; Sobotka, 2008), findings on the fertility of descendants of immigrants are mixed. Fertility among 1.5 and second generation women depends strongly on the origin group, parity and the national context considered (Andersson et al., 2017; Baek et al., 2025; Guarin Rojas et al., 2018; Höhn et al., 2024; Krapf and Wolf, 2015; Kulu and Hannemann, 2016; Kulu et al., 2017, 2019; Pailhé, 2017; Wilson, 2019).

To date, these two dimensions of social differentiation in family formation have not been systematically combined (Krapf et al., 2024; Marynissen et al., 2025). In the expanding literature on the fertility of migrants and their descendants, education is often included as a covariate in order to examine whether their fertility patterns become more (or less) similar to those of natives after controlling for educational differentials (Baek et al., 2025; Kulu et al., 2017; Pailhé, 2017). However, the intersection between education, origin group and migrant generation is rarely examined, leaving potential variation by migration background in the association between education and both entry into parenthood and parity progression largely unexplored (Krapf and Wolf, 2015). Similarly, research on educational gradients in the tempo and quantum of fertility has focused primarily on majority (or general) populations, while paying only limited attention to the question of whether similar associations hold across diverse origin groups and migrant generations (Krapf et al., 2024; Marynissen et al., 2025). Without examining how educational gradients vary across origin groups and migrant generations, we cannot assess whether education has an equalising effect on fertility behaviour across groups, or whether group-specific cultural and structural or contextual factors continue to shape childbearing even among individuals with similar educational levels.

Both the literature on migrant fertility and the literature on educational gradients in fertility face the challenge of how to account for selective entry into parenthood and subsequent selective progression to higher parities (Kravdal, 2001, 2007; Kreyenfeld, 2002; Wood et al., 2014). Unobserved characteristics that make some women more “prone to childbearing” than others may vary significantly across educational levels, as well as across origin groups and migrant generations. Accounting for such (differential) selection effects is therefore essential to avoid bias in the estimated relationship between education and fertility, and to enable meaningful comparisons across origin groups and migrant generations.

Building on our previous work that focused on entry into parenthood (Marynissen et al., 2025), this paper addresses these research gaps by examining the association between educational attainment, entry into parenthood and progression to second and third births among women of the 1.5 generation (who migrated before the age of 18) and the second generation in Belgium during the 2000–2011 period. Using population-wide register data, we distinguish six origin groups: women originating from (i) southern Europe, (ii) eastern Europe, (iii) northern and western Europe, (iv) Turkey, (v) Maghreb countries and (vi) other non-European countries, alongside native Belgian women (women without a migration background) as a reference group. We make three substantial contributions to the literature by answering the following three research questions: (1) To what degree do educational gradients in the timing and intensity of first, second and third births differ across origin groups and migrant generations? The extent to which educational gradients are similar across groups indicates whether education “transcends” other factors (e.g. cultural values, family networks) or is modified by specific cultural/structural contexts. (2) Conversely, to what degree do the timing and intensity of childbearing still vary across origin groups and migrant generations when considering women with the same level of education? The extent to which variation across origin groups and migrant generations persists when controlling for level of education indicates whether fertility differences are shaped primarily by factors beyond education (e.g. cultural values, family networks, economic conditions) or are mainly due to educational/socioeconomic factors. (3) Finally, to what degree do selective entry into parenthood and subsequent parities shape the education-fertility nexus across origin groups and migrant generations? In line with research on educational gradients in higher order births among majority populations, we address the issue of selectivity by modelling all three parity transitions jointly and including a time-constant shared frailty for recurrent events (i.e. random effect) at the level of individual women (Hougaard, 2000; Kravdal, 2001; Kreyenfeld, 2002; Wood et al., 2014). Building on our earlier work on the education-fertility nexus in migrant populations (Marynissen et al., 2025), the joint modelling of all three parity transitions now allows us to explore educational differences in entry into parenthood and progression to second and third births across origin groups and migrant generations simultaneously, net of (differential) selection over subsequent parities by comparing women with average frailties across groups. Furthermore, the variance estimates of the shared frailty term indicate whether and to what extent heterogeneity – for instance regarding fertility motivations, opportunities or restrictions – differs across origin groups and migrant generations. Answering these research questions can shed light on the larger issue of whether ongoing educational expansion will lead to a convergence of fertility patterns across diverse populations, or whether fertility differentials will persist due to structural and cultural differences, regardless of a convergence of educational attainment. Moreover, these insights can help to answer the question of whether continued educational expansion among migrant groups will contribute to further fertility postponement – as previously observed in majority populations – or whether other factors will counteract postponement trends, potentially stalling or reversing current patterns of delayed childbearing.

Belgium has a long history of immigration, with different ethnic minorities typically being rooted in different types of migration, which can, in turn, be linked to differential opportunity structures in terms of education and the labour market. Migrants from neighbouring northern and western European countries have often migrated to Belgium for reasons of education or employment. This has also been the case for more recent migrants from eastern Europe following EU enlargements (Myria, 2021). The fact that these migration channels are intrinsically connected to education and labour is often put forward as an explanation for the recurring finding that these migrants and their descendants have opportunity structures and labour market outcomes approximating those of the Belgian majority population.

In contrast, migrants and descendants of migrants originating from southern Europe, Turkey and Maghreb countries are more strongly rooted in the post-WWII labour migration of predominantly male and low educated workers. For southern European groups, this period of labour migration was followed by substantial levels of return migration due to economic growth in their origin countries and free movement within Europe since the 1960s. Thus, those southern European migrants who remained in Belgium or arrived later represent a more economically established and diverse population. More recent southern European migrants and later generations are now characterised by a more diverse profile in terms of their socioeconomic position and gender (Myria, 2016). In contrast, Turkish and Maghrebi labour migration was, in response to the migration stop in the early 1970s, followed by flows of family reunification and marriage migration that were largely unrelated to education or employment opportunities (Reniers, 1999). The concentration of Turkish and Maghrebi communities in urban industrial areas that experienced subsequent deindustrialisation has contributed to their persistent exposure to poverty and socioeconomic disadvantage (Costa and de Valk, 2018; Van der Bracht et al., 2014). Today, Turkish and Maghrebi migrants and their descendants, and particularly women, continue to be characterised by substantial disadvantages in terms of labour market participation (FOD WASO and Unia, 2022), which reflect both structural factors such as lower educational attainment and discrimination that creates additional barriers to economic integration (Van der Bracht et al., 2014; Wildan and Husein, 2021).

Finally, the group of migrants and their descendants from other non-European countries is highly diverse. This subgroup includes a substantial number of post-colonial migrants from Congo, as well as individuals originating from other (high-income) OECD countries, Asia, the US or South America. The origin countries have further diversified in recent migration flows, which often involve refugees and asylum seekers from, for example, Syria, Iraq, Afghanistan and Ukraine. The numbers of 1.5 and second generation migrants originating from single non-European origin countries are still small (making a further differentiation of the “other non-European” origin group in the analyses impossible), and these groups are highly diverse in terms of the ideational contexts and opportunity structures their members have experienced.

Whereas the 1.5 generation migrated during childhood or adolescence, and thus experienced partial socialisation in Belgian institutions, the second generation was born and entirely educated in Belgium. Nevertheless, as Table 1 shows, educational attainment varies substantially across origin groups for both generations. Educational expansion is evident across generations, with second generation women generally achieving higher educational levels than 1.5 generation women within most origin groups. However, the Turkish and Maghrebi groups continue to have lower educational attainment. In addition, educational credentials do not translate uniformly into economic opportunities across groups (Liebig et al., 2010). As a result, levels of labour market integration vary considerably, with discrimination and spatial concentration in deindustrialised areas creating additional barriers for some groups despite their educational gains (Costa and de Valk, 2018; Van der Bracht et al., 2014; Wildan and Husein, 2021). These differential opportunity structures suggest that the association between education and fertility – through mechanisms such as opportunity costs, economic resources or cultural adaptation – is likely to play out differently across origin groups and generations.

The mechanisms that link education to the timing and intensity of first, second and third births are complex and multifaceted, involving enrolment in education, labour market positions and opportunity costs, normative considerations, gender equality, attitudes and behaviour associated with higher levels of education and access to work-family reconciliation policies (Gustafsson et al., 2002; Robert-Bobée and Mazuy, 2005; Vasireddy et al., 2023; Wood, 2019). These mechanisms may operate differently across origin groups and migrant generations due to variation in cultural norms, migration histories, institutional contexts and structural constraints. Given the descriptive nature of this article, we do not explicitly examine these specific mechanisms or attempt to explain variation across origin groups and migrant generations through detailed causal pathways. Instead, we focus on a fundamental methodological challenge that affects any analysis of education-fertility associations, regardless of the underlying mechanisms: namely, how unobserved heterogeneity and selective entry into parenthood shape the observed associations between education, entry into parenthood and parity progression. Understanding and addressing this issue is essential for producing valid estimates of educational gradients in fertility, and for making meaningful comparisons across origin groups and migrant generations.

Unobserved characteristics of individual women – such as family orientation, aspirations, preferences or compliance with social norms, as well as more structural opportunities or constraints – have been shown to influence both the timing and intensity of entry into parenthood, as well as subsequent parity progression. Controlling for such selection mechanisms is essential when comparing groups since women entering parenthood at younger ages tend to have a specific profile (reflected in a higher risk or frailty of having a child), which, in turn, leads to a selective group of women entering the risk set for second and higher order births (Kravdal, 2001). Likewise, the group of women still at risk of having a first child will become increasingly selective with age (characterised by a lower individual risk or frailty), leading to deflated conditional first birth probabilities at these ages (Kravdal, 2001; Vaupel et al., 1979; Vaupel and Missov, 2014). This pattern holds for women across all educational levels. However, higher educated women generally start childbearing at later ages due to, among other factors, their longer enrolment in education and higher employment probabilities and wage potential compared to lower educated women. This implies that, for example, low educated women with high frailty may enter parenthood at around age 22 and those with low frailty may enter parenthood at around age 28, while higher educated women with high frailty may enter parenthood at around age 28 and those with low frailty may enter parenthood at around age 34. Comparing conditional first birth probabilities between low and higher educated women at age 28 without controlling for this unobserved heterogeneity would effectively compare low educated women with low frailty to higher educated women with high frailty. This would result in an underestimation of the conditional first birth probabilities of low educated women, an overestimation of the conditional first birth probabilities of higher educated women and an overestimation of the difference between the two (at age 28).

Furthermore, selective entry into parenthood shapes the composition of the risk set for second and higher order births (Kravdal, 2001). Since women with higher frailties are more likely to have had a first birth, the risk set for second births becomes positively selected (i.e. it consists disproportionally of women with higher frailties), inflating conditional second birth probabilities. This selection strengthens with parity, making the estimation of second and third birth probabilities increasingly prone to bias if unobserved heterogeneity is ignored. Moreover, as the strength and structure of this selection process have been shown to differ by level of education, estimates of the association between education and fertility may be substantially biased. Incorporating a shared frailty for recurrent events thus avoids conflating the effect of education on transitions to higher parities with selective entry into parenthood or earlier parities.

Controlling for selection is particularly important when comparing fertility patterns across origin groups and migrant generations, as the influence of unobserved factors on childbearing decisions may vary substantially across groups, resulting in different variance estimates for the individual-level random effects or frailties, and different selection processes across parities. Such unobserved factors may range from cultural norms regarding ideal family size and timing to religiosity, gender role attitudes, intergenerational expectations and constraints related to discrimination or access to resources – many of which are difficult or impossible to measure directly. If these unobserved characteristics are systematically related to fertility behaviour, and their distribution differs by origin group and generation, failing to account for them will not only induce bias in the estimated educational gradients of entry into parenthood and parity progression, but will also affect the comparison of fertility patterns across origin groups and migrant generations.

In what follows, we examine variation in the education-fertility link across origin groups and migrant generations, separately for each parity. In line with our research questions, for each parity, we first explore educational gradients within origin groups and migrant generations to examine whether the effect of education is similar across groups or is moderated by cultural and/or structural factors specific to each group (RQ1). We then turn to variation in the timing and intensity of childbearing across origin groups (and migrant generations) within educational levels to study to what extent fertility differences across origin groups can be attributed to educational/socioeconomic factors (RQ2). Finally, we compare the three variants of the coefficients of variation to assess the role of selective entry into parenthood in shaping the observed associations (RQ3).

Fertility transitions in contemporary European societies are characterised by pronounced social differentiation along multiple dimensions. While research has documented variation by both migration background and education, these dimensions have not been systematically combined to investigate how they jointly shape the tempo and quantum of fertility. This study addressed this gap by examining the intersection of educational attainment, origin group and migrant generation in shaping entry into parenthood and progression to second and third births among women in Belgium in the 2000–2011 period. Using population-wide register data and shared frailty models that account for selective entry into parenthood, we addressed three interconnected research questions that reveal the interplay between education and migration background in fertility transitions.

Our first research question investigated to what degree educational gradients in the timing and intensity of first, second and third births differ within origin groups and migrant generations. The results reveal striking differences in how education shapes fertility across groups and birth orders. For first births, clear educational gradients emerge across all origin groups and migrant generations: the life table proportion of women having a first birth (SPPR1) decreases with higher levels of education, while the mean ages at first birth (SMAC1) increase. However, the strength of this gradient varies considerably, with more pronounced variation being observed among women of southern European, eastern European, Turkish and Maghrebi origin than among women of Belgian, northern and western European and other non-European origin. For second births, educational gradients become highly group-specific. Among most origin groups and migrant generations, higher education is associated with shorter intervals between the first and the second birth (as reflected in lower mean birth intervals, SMBI2). In contrast, the life table proportion of women progressing to a second birth (SPPR2) increases with higher education only among women of Belgian origin, second generation women of southern, eastern, northern and western European origin and 1.5 generation women of northern and western European origin. In all other origin-generation groups, no such educational gradients in SPPR2 emerge, suggesting that other (structural or cultural factors) override educational influences on the decision to have a second child. This divergence likely reflects different mechanisms operating across groups. Among women of Belgian and European origin, higher education may facilitate the transition to a second birth by providing them with greater economic resources and access to work-family reconciliation policies (Biegel et al., 2021; Marynissen et al., 2021; Wood, 2025). In contrast, the more limited economic returns to education for non-European origin groups may, potentially in combination with cultural norms favouring multi-child families (Milewski and Mussino, 2018), result in a decoupling of education and higher order fertility (Kreyenfeld and Andersson, 2014; Wood and Neels, 2017). For third births, our findings show no clear or consistent educational gradients, despite substantial variation in the life table proportions of women progressing to a third birth (SPPR3) and in the mean birth intervals (SMBI3) across educational levels within origin groups and migrant generations. The patterns are diffuse and unpredictable. As such, higher education does not appear to contribute to a convergence across origin groups in the likelihood of transitioning to a third birth, indicating that at higher parities, factors beyond educational attainment become increasingly decisive in shaping fertility.

Our second research question examined to what degree the timing and intensity of childbearing vary across origin groups and migrant generations when educational attainment is held constant. Our findings reveal that substantial variation persists within educational levels, although the extent varies by birth order. For first births, variation in synthetic parity progression ratios to a first birth (SPPR1) and synthetic mean ages at first birth (SMAC1) across origin groups is lowest among the highly educated and the second generation. This suggests that higher education is associated with a convergence in both the timing and intensity of first births across origin groups, particularly among the second generation. This finding is consistent with previous research on the education-parenthood link among 1.5 and second generation migrants (Krapf and Wolf, 2015; Marynissen et al., 2025; Pailhé, 2017), suggesting that education has an equalising effect on entry into parenthood – be it through more similar employment probabilities, access to work-family reconciliation policies, more egalitarian gender role attitudes and/or more similar socialising experiences and normative expectations associated with higher education. For second births, there is considerable variation across origin groups, but it again tends to be smaller among highly educated women, particularly for SPPR2. There is no consistent pattern of lower variation among the second generation than among the 1.5 generation. These findings suggest that while higher education fosters convergence in the timing and intensity of second births across origin groups, this convergence remains more limited than it is for first births. For third births, variation across origin groups within educational levels remains substantial among all educational groups, indicating that factors beyond educational attainment may play a more decisive role at higher parities.

Our third research question examined to what degree selective entry into parenthood and subsequent parities shapes the education-fertility link across origin groups and migrant generations. The shared frailty models reveal moderate to substantial unobserved heterogeneity in childbearing propensity across all origin groups and migrant generations. This heterogeneity is consistently higher among the 1.5 generation than among the second generation, and – with the exception of second generation women of Turkish origin and women of Maghrebi origin – is higher among all origin groups and migrant generations than among women of Belgian origin. These patterns suggest that migrant populations are characterised by greater diversity in unmeasured factors affecting fertility behaviour, which differentially affects the progression to first, second and third births across groups. These results underscore the necessity of taking unobserved heterogeneity explicitly into account in future work on the association between education and parity progression across diverse population groups.

Finally, this study has several limitations that point to fruitful paths for future research. First, despite the use of population-wide register data and a modelling strategy that addresses unobserved heterogeneity, several data limitations and analytical choices may affect our findings. Education is measured as a time-constant indicator, meaning that the estimated effects of education conflate how education influences fertility with how fertility may affect later educational attainment (Kravdal, 2001). In addition, the desire to make more detailed distinctions by origin group and generation is traded off against the need to obtain stable and interpretable results on variation across origin groups and generations in the association between education, entry into parenthood and parity progression (i.e. results that are not too erratic due to small cell counts when combining all these dimensions). As a consequence, intra-group differences may be obscured by aggregating origin groups into rather broad categories, and by not differentiating the group of child migrants (1.5 generation) further by age at migration – which is an essential factor in integration and exposure to host country institutions, including the educational system. Furthermore, while estimating the models over an 11-year window (2000–2011) enables better identification of unobserved heterogeneity, it limits the ability to detect changes over time, timing-sensitive developments (e.g. policy changes or economic shocks) as well as more recent changes in migration flows. In particular, the analyses do not consider the substantial immigration waves after 2011, including increasing numbers of refugees and asylum seekers, which primarily altered the compositional profile of the 1.5 generation in terms of origin, migration motives and socioeconomic background. To the extent that post-2011 migrant cohorts include a higher share of refugees and migrants originating from contexts characterised by disrupted schooling or constrained labour market opportunities, the association between educational attainment and fertility timing observed in this study may be weaker or operate through different mechanisms among these cohorts. At the same time, increased educational integration efforts and policy attention to migrant children in more recent periods may counteract some of these effects. Consequently, caution is warranted when extrapolating the observed associations between education, entry into parenthood and parity progression to more recent cohorts of migrants – particularly those of the 1.5 generation. While shared frailty models are often used to capture unobserved heterogeneity, the assumption of a time-constant and normally distributed random effect (with the same structure across origin groups and migrant generations) may oversimplify real-life variation. In sum, future research would benefit from the inclusion of data on more recent periods, time-varying information on educational enrolment, more detailed distinctions between origin groups and migrant generations (if cell counts allow) and more flexible modelling of unobserved heterogeneity. Second, as a result of the descriptive nature of the paper, the potential mechanisms that link education to entry into parenthood and parity progression have not been explicitly considered in the model or empirically tested. Future research could include information on partnership status, employment status, income, partner characteristics or access to work-family reconciliation policies to enhance the understanding of how structural and normative factors shape the association between education and fertility patterns across origin groups and migrant generations.