The available literature on migrant fertility highlights that (dis)similarities in the fertility patterns of migrants and natives may be driven not only by variation in norms and preferences, but also by variation in (responses to) the opportunity structures and institutional settings of the host society (Andersson and Scott, 2005; Kulu and González-Ferrer, 2014; Kulu et al., 2017, 2019). These factors are likely to differ across migrant groups with different migration histories, which may, in turn, yield variation in the nexus between education and first births.

Belgium has a sizeable and diverse migrant population, the largest share of which originates from neighbouring countries in Northern and Western Europe that have fertility levels similar to those in Belgium. A large proportion of this group moved to Belgium for reasons of education or work, rather than union formation (Myria, 2021), and are quite similar to the Belgian majority population in terms of the opportunity structures and labour market outcomes they experience. Other large migrant groups originate from Southern European countries. ² This origin group has similarities with a number of non-European origin groups (i.e., Turkey and Maghreb countries), as they are all rooted in the labour migration after WWII. However, these Southern European groups differ considerably in terms of their subsequent patterns of migration, as well as in terms of the socio-economic and ideational contexts in which the 1.5 and second generations grew up. Due to economic growth in their origin countries and the free movement within Europe since the 1960s, return migration was common among the predominantly male and low educated Southern European labour migrants. In contrast, Southern European migrants who arrived after 1980 (when Southern European countries first started to experience very low fertility levels) have a more diverse profile in terms of their socio-economic position and gender (Myria, 2016). Today, this migrant group holds an intermediate position regarding labour force participation, as they have fewer labour market opportunities and lower employment rates than the Belgian majority population, but more favourable opportunity structures compared to, for example, Maghrebi and Turkish migrant groups (FOD WASO and Unia, 2022). Migrant groups and generations originating from Eastern European countries, ³ which are generally low fertility contexts (OECD, 2021), did not become more numerous until the 1990s onwards. Up to the present, the migration of these groups is largely driven by employment opportunities and – for migrants from Eastern European countries that are part of the European Union – free movement within the European Union.

Migrant groups from Maghreb countries ⁴ and Turkey also originated from post-WWII labour migration flows, but following the migration stop in the early 1970s, family reunification and family migration (e.g., marriage migration) became (and still are) important migration motives in these migrant groups. While migration and labour market participation were closely linked among Turkish and Moroccan men who migrated in the context of labour migration, the migration of their female partners was not related to employment. This may have affected the labour market opportunities for first generation Turkish and Moroccan women (Surkyn and Reniers, 1997), and might have fostered more traditional gender patterns among groups originating from Turkey and Morocco, which have higher fertility levels than those of Belgium. Women from both Maghrebi and Turkish migrant groups – particularly women of the first generation, but also women of the second generation – are overrepresented in precarious employment, unemployment and inactivity (FOD WASO and Unia, 2022). Finally, migration from other non-European countries has also contributed to Belgium’s migrant population, particularly as a result of post-colonial migration from Congo, but the origin countries have diversified in recent migration flows (often refugees and asylum seekers). This group is very diverse in terms of the ideational contexts and opportunity structures its members experience, and its 1.5 and second generations are still relatively small. This group is more strongly represented in more recent migration waves from, for example, Syria, Iraq and Afghanistan.

A large body of literature on the association between education and entry into parenthood has shown that higher educational attainment is associated with postponement of first births, and, in case of partial recuperation at older ages, higher childlessness, due to five underlying mechanisms. Closely related to the fact that the aforementioned origin groups come from specific migration contexts, we argue that these mechanisms are likely to vary by migration background.

First, research has shown repeatedly that first birth rates are substantially lower during periods of enrolment in education (Blossfeld and Huinink, 1991; Neels et al., 2017). Among the proposed explanations for this pattern are that full-time enrolment in education is incompatible with family responsibilities (Cohen et al., 2011; Gustafsson et al., 2002), and that social norms regarding the sequencing of life course transitions in young adulthood – i.e., finishing education, securing employment and financial independence, union and family formation – discourage childbearing while in education (Oppenheimer, 1994; Robert-Bobée and Mazuy, 2005). As such, higher levels of education are expected to give rise to postponement of parenthood because they require extended enrolment in education. The interplay between enrolment and entry into parenthood can be expected to play out somewhat differently, however, by migration background. Turkish and Maghrebi origin groups in particular exhibit more discontinuities in educational trajectories leading to early drop-out, as well as longer enrolment in education due to grade repeating (Neels, 2000). This implies that the number of years of schooling needed to achieve a given level of education and the length of educational trajectories can be expected to vary more in these origin groups, ranging from short periods of enrolment for women who drop out early and thus have low levels of education, to relatively long educational careers for a selective group of women who attain higher educational levels – albeit with higher probabilities of experiencing discontinuities and grade-repeating during their trajectories.

Second, higher levels of educational attainment are associated with higher employment probabilities and wage potential. Given the importance of early career investments for gaining access to long-term career tracks with favourable wages and working conditions (e.g., autonomy, flexibility), family formation shortly after graduation comes with high opportunity costs for highly educated women (Gustafsson et al., 2002), encouraging further postponement of parenthood to older ages (Liefbroer and Corijn, 1999). In case of only partial recuperation of fertility later in the life course, higher levels of education may also entail higher childlessness. When considering differentials by migration background, we expect that the variation in opportunity costs by level of education is more articulated, particularly among origin groups with limited labour market opportunities. The well-documented structurally limited labour market opportunities for women of foreign – and particularly non-European – origin are likely to discourage these women from investing in labour market trajectories, and to encourage them to take up alternative roles that are socially valued, such as motherhood. In contrast, we assume that the small and highly selective group of women with a migration background who attain high levels of education will also be confronted with the high opportunity costs associated with early childbearing. However, in line with micro-economic theories arguing that time investments are path-dependent as they become increasingly efficient, the select group of highly educated women is expected to postpone childbearing more strongly than native women, as they have invested in career development, but are subsequently confronted with persistent barriers in the labour market (e.g., discrimination in combination with sticky floors or glass ceilings).

Third, work-family reconciliation policies potentially reduce the opportunity costs of childrearing. Given that the opportunity costs of childrearing are most elevated for highly educated women, it is likely that these women will benefit most from policies such as formal childcare or parental leave when seeking to realise their fertility intentions while maximising labour market returns to their educational credentials (Pavolini and Van Lancker, 2018; Van Lancker and Ghysels, 2012; Wood et al., 2023). In addition, policy design features such as high costs of formal childcare or strict employment-related eligibility criteria for parental leave may further exacerbate differential benefits from work-family policies by level of education (Abrassart and Bonoli, 2015; Biegel et al., 2021; Kil et al., 2018; Marynissen et al., 2021; Viitanen, 2005). The available literature has repeatedly shown that there are ethnic gaps in the uptake of formal childcare, with the uptake being particularly low among women with a non-European migration background. These gaps have been linked to a wide range of barriers on both the demand side (e.g., lack of employment stability) and the supply side (e.g., design features) affecting women with a migration background (Biegel et al., 2021). To the extent that such barriers also reduce access to family policies for highly educated women with a migration background, the role of formal childcare in work-family reconciliation may be weakened, further increasing the opportunity costs of childrearing.

Fourth, higher levels of gender equality in the division of unpaid labour within households (e.g., childrearing or household tasks) reduce the opportunity costs of childbearing for women. Allowing for variation across countries, the available literature has repeatedly shown that highly educated women are more likely to experience egalitarian gender role attitudes and divisions of labour within the household (Evertsson et al., 2009; Phinney and Flores, 2002). As such, variation in work-family reconciliation strategies by level of education (e.g., uptake of family policies and egalitarian gender roles within households) are expected to limit postponement and support recuperation of first births among highly educated women. To the extent that highly educated women from different migration backgrounds also face different ideational contexts and contributions to unpaid labour from their male partners, variation in the postponement and recuperation of first births can be expected. The available literature is inconclusive, however, regarding variation in gender norms by migration background. On the one hand, the cultural and institutional climates in the country of origin are geared more towards the male breadwinner model for several origin groups, and parental attitudes and family networks may be expected to act as a source of origin-specific values for women of the 1.5 and second generations (Idema and Phalet, 2007; Marks et al., 2009). On the other hand, women of the 1.5 and second generations do not have any direct experience of their parents’ origin country, and empirical evidence shows substantial heterogeneity in gender role attitudes across and within origin groups (Wood, 2022), with migrant-native differentials strongly depending on gender (de Valk, 2008).

Fifth, ideational explanations of fertility patterns suggest that ideational changes regarding the family and the combination of work and family that accompany educational expansion have given rise to shifting norms regarding the timing of childbearing among higher educated women (Gustafsson et al., 2002; Lesthaeghe and Surkyn, 1988; Neels et al., 2017; Van de Kaa and Lesthaeghe, 1986). To the extent that higher levels of education are associated with a stronger emphasis on individual rather than family-related goals, such values can be expected to become drivers of fertility postponement, or even childlessness, among the higher educated in their own right. Depending on whether higher education among women with a migration background is accompanied by values that support individual self-realisation and postponement of family formation to safeguard their labour market trajectories, or whether these women remain more exposed to alternative ideational frameworks that prescribe other social roles for women, additional variation by migration background in patterns of postponement and recuperation of fertility by level of education can be expected to emerge.

This paper uses population-wide retrospective longitudinal microdata drawn from the Belgian census of 1 January 2011. This was the first census that was fully register-based, and it provides information on country of birth, citizenship at birth, year of immigration and descent. The information on descent links parents to children, and is vital to determine women’s migration backgrounds and to reconstruct their maternity histories. The 2011 census was validated against vital registration and provides quasi exact retrospective estimates of the period total fertility rate (period TFR) between 1985 and 2010. As the period TFR is sensitive to shifts in the timing of fertility (Bongaarts and Feeney, 1998), and cohort fertility indicators cannot yet be calculated for origin groups with more recent migration histories, this study adopts a synthetic (period) life table approach to compare the timing of a first birth and the proportion of women entering parenthood across origin groups and migrant generations (Feeney and Yu, 1987; Neels, 2006; Ni Bhrolchain, 1987). ⁵ To this end, we observe childless women of Belgian origin and women with a migration background who were exposed to the risk of having their first child between ages 18 and 50 in the 1990–2010 period, thus contributing 16,873,755 person-years of exposure to the construction of the synthetic life tables of entry into parenthood.

An individual’s origin group is determined using information on the citizenship at birth of one’s parents, and the citizenship at birth of women themselves. ⁶ A woman’s origin is considered to be the citizenship at birth of her mother. If this information is missing, or if the citizenship at birth of a woman’s mother is Belgian, we consider her father’s citizenship at birth to reflect her origin. Subsequently, if information on the citizenship at birth of a woman’s father is missing or if it is Belgian, we consider the woman’s citizenship at birth to reflect her origin. If there is no information on the citizenship at birth of a woman and her parents, the woman’s current citizenship is considered to reflect her origin. ⁷ We distinguish seven origin groups: Belgium (BE), Southern Europe (SEU), Eastern Europe (EEU), Northern and Western Europe (NWEU), Turkey (TKY), Maghreb countries (MGB) and other non-European countries (otNEU).

Migrant generation is determined based on women’s origin, country of birth and age at migration. If women have a non-Belgian origin, were not born in Belgium and migrated to Belgium before age 18, they are considered to be part of the 1.5 generation (1.5G). If women have a non-Belgian migration background, but were born in Belgium, they are considered to be part of the second generation (2G).

Women’s highest level of education is determined using the information on their level of education in the 2011 census, complemented with follow-up register data on level of education up to 2017. ⁸ Due to the retrospective design, we have no time-varying information on enrolment and level of education over the life course. The use of a time-constant indicator on the level of education obtained by the 2011 census can introduce bias into the association between education and fertility, however, as the temporal ordering of events – leaving education versus having a first child – cannot be identified (Hoem and Kreyenfeld, 2006). ⁹ This limitation should therefore be taken into account when interpreting the association between education and first births documented in this study. The education variable distinguishes three educational groups based on the ISCED 1997 classification: (1) low educated women with no education, primary or lower secondary education (ISCED 0, 1 and 2); (2) medium educated women with higher secondary and post-secondary non-tertiary education (ISCED 3, 3A, 3C and 4); and (3) high educated women with short and long forms of tertiary education (ISCED 5, 5A, 5B and 6). Table 1 shows the distribution of level of education by origin group and migrant generation, as well as the observed synthetic parity progression ratio to a first birth (SPPR1) and the associated synthetic mean age at first birth (SMAC1). Table S.1 in the supplementary material (available online at https://doi.org/10.1553/p-mf52-3j2z) additionally documents the distribution of person-years by age (five-year categories), level of education, origin group and migrant generation.

We model entry into parenthood by following childless women from the age of 18 up to and including the year in which their first child was born, or when censoring occurs as a result of reaching the age of 50 or the end of the observation period on 31 December 2010. ¹⁰ Because of the retrospective design, no censoring occurs due to emigration or death.

We use discrete-time hazard models with a complementary log-log link function to model women’s conditional probability of having a first birth q ( a ) i t at age a . We estimate three models to examine variation by migrant origin and generation in the education-fertility nexus. These models gradually build up to test whether, in addition to variation in the education-fertility nexus by migrant origin, there is also significant variation by migrant generation. Model 1 includes age ( A t i ) in years (period difference) centred at the age of 18 (baseline, fourth order polynomial), ¹¹ education ( E i ), origin group ( O i ), the interactions between education, origin group and the linear and quadratic terms of the baseline, and period ( P t i ). We include this categorical variable for period to control for contextual factors that have not explicitly been taken into account in the analyses, as well as compositional differences between origin groups and generations throughout the observation period. This variable distinguishes between periods 1990–1996, 1997–2003 and 2004–2010. Model 2 additionally includes migrant generation ( G i ). Finally, Model 3 relaxes the proportionality assumption regarding migrant generation by allowing the baseline to vary not only by level of education and origin group, but also by migrant generation (including the four-way interaction in Equation (1)): − ln [ 1 − q ^ ( a ) i t ] = ∫ t t + 1 h ( a ) d a = e α ^ · e F ^ ( A t i ) · e F ^ ( E i ) · e F ^ ( O i ) · e F ^ ( G i ) · e F ^ ( A t i * E i * O i * G i ) · e F ^ ( P t i ) (1)

Model comparison indicates that including the four-way interaction between the baseline (age), level of education, origin group and migrant generation in Model 3 entails a significant improvement in model fit (likelihood ratio test with chi 2 ( 36 ) = 96,177 and p < 0.001) over Model 2 (which estimated first birth hazards as a function of the baseline, level of education, period, the three-way interaction between the baseline, level of education and origin group, only controlling for migrant generation). The model for women with a Belgian origin was run separately and includes age in years (period difference) centred at the age of 18 (baseline, fourth order polynomial), education, and the interaction between education and the linear and quadratic terms of the baseline. Table S.2 in the supplementary material provides an overview of the different model specifications.

Using the predicted conditional probabilities of having a first birth generated by these models, we plot first birth schedules by origin group, migrant generation and level of education. Furthermore, consistent with previous research (Neels et al., 2024; Wood et al., 2014), we use estimated first birth hazards to calculate the synthetic parity progression ratio to a first birth (SPPR1) and the corresponding synthetic mean age at first birth (SMAC1) by origin group, migrant generation and level of education (Equations (2) and (3)), in view of comparing educational differentials in postponement, recuperation and final intensity of first births across origin groups and migrant generations: SPPR 1 = 1 − ∏ a = 18 49 [ 1 − q ^ ( a ) ] (2) SMAC 1 = ∑ a = 18 49 a ( S a − S a + 1 ) SPPR 1 (3)

The synthetic parity progression ratio to a first birth is the final intensity at age 49 in the period attrition table for progression to a first birth, thus reflecting the life table proportion of women having a first child in the period considered (Feeney and Yu, 1987; Ni Bhrolchain, 1987). Indicators of order-specific fertility drawn from synthetic life tables solely rely on rates of the first kind in their calculation (where women who already had their first child are no longer included in the risk set) and have been shown to provide close approximations of cohort fertility compared to indicators based on rates of the second kind (such as order-specific period total fertility rates), while using the most recent information available (Calot, 2001; Neels, 2006).

Figures 1(a)–1(f) show the estimated conditional first birth probabilities from Model 3 by age and origin group for the six combinations of educational attainment (low, medium, high) and migrant generation (1.5G versus 2G). Table 2 reports the corresponding synthetic parity progression ratios to a first birth (SPPR1, Panel 1) and the synthetic mean ages at first birth (SMAC1, Panel 2) by migrant generation, origin group and level of education, which are also represented graphically in Figure 2. Coefficients of variation (standard deviation divided by the mean) were additionally reported in Table 2 to measure variation in SPPR1 and SMAC1 across origin groups (both overall and within ISCED levels), as well as variation in SPPR1 and SMAC1 across ISCED levels within origin groups.

As expected and in line with previous literature, Figures 1(a)–1(f) show that the schedules of conditional first birth probabilities in all groups consistently shift to older ages with increasing levels of education, while the first birth schedules of women with a migration background also show variation around the schedule of native women with the same level of education (the black line in Figures 1(a)–1(f)), giving rise to variation in SMAC1 and SPPR1 across origin groups and migrant generations compared to native women (Table 2 and Figure 2). A fairly consistent pattern emerges with respect to educational differentials in the synthetic mean age at first birth: the mean age at first birth increases with higher levels of education, and although variation across origin groups in the timing of the first birth is smaller when considering women with the same level of education (CVar(origin), Table 2), educational differentials in the mean age at first birth are typically somewhat more articulated among women with a migration background than among natives, particularly in the 1.5 generation (CVar(isced), Table 2). The picture is less uniform, however, when considering educational gradients in the life table proportion of women having a first child: whereas the educational gradient has virtually disappeared among native women, pronounced negative gradients do emerge in several origin groups for both the 1.5 and the second generation.

The largest variation in both the timing and the proportion of women having a first child emerges among women of the 1.5 generation (Figures 1(a), 1(c) and 1(e)), and particularly among lower educated women. Lower educated women of the Turkish and Maghrebi 1.5 generation show high first birth hazards at relatively young ages, resulting in lower SMAC1 (resp. 21.15 and 22.23), but also higher SPPR1 (resp. 0.927 and 0.933), compared to lower educated women of Belgian origin (SPPR1 of 0.821 and SMAC1 of 23.01). Lower educated women originating from Eastern Europe also have an earlier fertility schedule than lower educated native women, and a somewhat higher proportion of these women have a first child, whereas lower educated women of Southern European origin combine somewhat later fertility with higher synthetic parity progression ratios to a first birth compared to lower educated natives. Finally, lower educated women of the 1.5 generation of Northern and Western European origin as well as of other non-European origin combine later fertility schedules with somewhat higher synthetic parity progression ratios to a first child compared to lower educated natives. Compared to lower educated women, variation in first birth timing across origin groups is more limited among medium educated women of the 1.5 generation, with the proportion of women having a first child being similar to that among native women, with the exception of women of Turkish and Maghrebi origin. This pattern is even more articulated for higher educated women, among whom all origin groups have similar or later mean ages at first birth compared to those of native women, albeit with lower proportions of women having a first child (again, with the exception or women of Maghrebi origin, among whom the proportion is similar to that of natives).

Among women of the second generation, variation in the timing of a first birth across origin groups is limited when considering women with the same level of education, as the highest first birth hazards are located at similar ages in all origin groups (Figures 1(b), 1(d) and 1(f)). Variation across origin groups is more substantial, however, when considering the life table proportion of women having a first child. Among lower and medium educated women, only second generation women of Turkish origin stand out as having somewhat higher first birth hazards at younger ages than other origin groups, which are combined with a higher synthetic parity progression ratio to a first birth. Moreover, among lower and medium educated second generation women of Southern European, Northern and Western European and Maghreb origin, somewhat higher percentages have a first child than among lower and medium educated native women. However, when considering higher educated women in different origin groups, we see that second generation women consistently have lower synthetic parity progression ratios to a first birth compared to native women.

A comparison of the cumulative fertility schedules of medium and higher educated women to that of lower educated women in each of the origin groups indicates that postponement of first births with increasing education is generally more pronounced among migrant origin women than it is among native women. In addition, there is also clear variation across migrant generations, with the educational differentials in the timing of a first birth typically being largest among women of the 1.5 generation, whereas the differentials among second generation women are more similar to the patterns found among natives.

A comparison of the educational differentials in the postponement of fertility across origin groups indicates that women of Western European and Northern European origin have educational differentials in the timing of parenthood that are most similar to those among natives, followed by women of other non-European origin. In contrast, education is associated with stronger differentiation in the timing of parenthood compared to that among native women (i.e., stronger postponement with increasing education) among women of Southern and Eastern European origin, and particularly among women of Turkish and Maghrebi origin, where the educational differentials in the timing of a first birth are most articulated for women of the 1.5 generation, but diminish and converge to the pattern among native women for women of the second generation. This pattern of convergence clearly occurs in the Turkish and Maghrebi origin groups, with postponement being less articulated among medium and highly educated women of the second generation than of the 1.5 generation. Similar results are found in the Eastern European origin group, where the strong educational differentials observed in the 1.5 generation do not occur in the second generation. Similarly, within groups originating from Southern Europe, postponement of the first birth with increasing education is more prevalent among the 1.5 generation than among native women, whereas the fertility postponement pattern among the second generation more closely approximates that among highly educated women in the Belgian origin group.

Within the Belgian origin group, the deficits at younger ages in the proportions of medium and highly educated women who already have their first child converge to zero by the age of 40, indicating that the postponement of the first birth with higher levels of education is (quasi) fully recuperated at older ages. This is not the case in the majority of other origin groups: a comparison of the recuperation of first births and the final intensities among medium and highly educated women vis-à-vis their lower educated age mates shows articulated negative educational gradients in the proportion of women having a first child, which vary strongly across the origin groups and migrant generations considered.

The patterns of education-related postponement and recuperation among women from the Northern and Western European origin groups vary little from those of native women, resulting in small differences in SPPR1 compared to natives, and a virtually neutral educational gradient in the proportion of women having a first child. Also 1.5 and second generation women originating from other non-European countries show limited educational gradients in the synthetic parity progression ratio to a first birth by the age of 35, with the fertility of higher educated women typically being somewhat lower than that among native women. In contrast, the results for women of Southern European origin indicate that the levels of recuperation are partial among medium and highly educated women of the 1.5 generation, resulting in these women having lower first birth intensities at age 40 than their lower educated age mates. Among second generation women of Southern European origin, however, the patterns of fertility postponement and recuperation at older ages are more similar to those among native women, resulting in a decline of the negative educational gradient in the proportion of women having a first child. Similar findings are observed for the Turkish origin group: educational differentials in the postponement and recuperation of first births among second generation women more closely approximate the patterns among native women than those among women of the 1.5 generation. Although medium educated women in the 1.5 and second generations still manage to catch up to lower educated women, a substantial deficit of first births relative to the levels among lower educated women persists among higher educated women of both the 1.5 and the second generation. A similar pattern emerges among higher educated women of Maghrebi and Eastern European origin: education-related postponement among the 1.5 and second generations is not compensated for by recuperation at higher ages, resulting in stronger negative educational gradients in the proportion of women having a first child compared to those among the Belgian origin group. Unlike in the Southern European origin groups, the fertility deficit of higher educated women compared to lower educated women is not smaller in the second generation than in the 1.5 generation in the Turkish, Maghrebi and Eastern European origin groups.