Past research has established that fertility rates among women differ by their level of education (e.g., Rindfuss and Bumpass 1980; Kravdal 2002; Lutz and Skirbekk 2014) and by their country of birth (e.g., Sobotka 2008; Kulu et al. 2019). How important is it to take these differentials into account when projecting the population of the EU? To what extent does using different approaches to modelling fertility affect the projected number of births? Both the expansion of education and the immigration of foreign nationals into the EU are trends that are likely to continue in the future, and that will play a role in shaping the population of the EU.

We argue that the multi-dimensional modelling of fertility is needed to capture the implications of differences in fertility across socio-economic groups, as the EU population is becoming increasingly diverse as a result of higher and more diversified migration flows in many EU countries in recent decades. When projecting population, it is important to account for immigration from third countries with generally higher fertility levels than the host EU countries, because it impacts the composition of the female population of reproductive age (by nativity, and also by education) – and, consequently, the number and the structure of births, as well as the size and the composition of the population. As differences in fertility between the native-born population and various groups of immigrants are partly associated with educational attainment (e.g., Roig Vila and Castro-Martín 2007), it is important to capture both dimensions simultaneously. Building on the well-established knowledge about differentials in fertility by education and migration status, we employ microsimulation modelling to demonstrate the effect of accounting for such differences in population projections. We focus on differences in the projected total fertility rate (TFR) and in the number of births and their composition. We present several scenarios that differ in how they model fertility and in the migration assumptions they use. As this work has been developed within the CEPAM project,¹ it focuses on EU28 countries.

Conventionally, multistate population projection methods (Rogers 1980, 1986; Schoen 1988) are used to take into account population diversity beyond the standard components of age and sex. The dimensions that are usually included in multistate projections influence population dynamics, and are associated with differences in fertility, mortality, and the propensity to migrate, such as educational attainment, which is closely linked to all components of population change (Lutz and Goujon 2001; Lutz et al. 2014). Among the other dimensions that are commonly included in such projections are place of birth (Rogers et al. 1999), religion (Goujon et al. 2007; Stonawski et al. 2015), ethnicity (Rees 2002; Raymer et al. 2018), and race (Colby and Ortman 2015). Multistate population projections usually include one, or at most two dimensions in addition to age and sex because adding three or more dimensions at the same time increases the complexity of this type of model to the point where it is difficult to handle any additional state. In a full multistate model, each additional state multiplies the set of transitions between the states. Data limitations also play an important role, particularly when the projection is for a group of countries with differing levels of data quality.

In contrast, microsimulation modelling offers a flexible way of taking into account a larger number of characteristics simultaneously. It is easier to handle a large number of parameters, because adding a dimension does not require multiplying all of the others. Adding a dimension simply involves adding a single column in the database. Among the advantages of microsimulation models are that they can provide richer outputs, and they can be used to build individual-level determinants of events. Thus, such models can provide better results when the number of dimensions included is relatively large, and when handling interactions between multiple variables (van Imhoff and Post 1998). The CEPAM-mic microsimulation model captures both the educational and the migrant characteristics of individuals, and its dimensions include age, sex, educational attainment, region of origin, duration of stay in the host country, and several others detailed in section 3.2. Importantly, microsimulation facilitates the appropriate modelling of period fertility because it enables us to include (rather than to adjust for) tempo effects in fertility that arise due to educational enrolment and the interplay between the migration process and childbearing.

This paper is structured as follows. As a backdrop to the modelling of fertility, we summarise the most enduring findings on fertility differentials by the educational and the migrant characteristics of women. In the data and methods section, we explain our general approach of contrasting the conventional multistate method with the microsimulation, while specifically focusing on the approaches to and the challenges involved in modelling fertility differentials. We then introduce the projection scenarios. In the data section, we also describe the CEPAM-mic microsimulation model, including its modules, its baseline population estimates, and its parameters. In a dedicated subsection, we present in detail the fertility module, including the data and estimation procedures for the parameters, and the results obtained using logit regression on Labour Force Survey data. In the results section, we focus on the compositional effects of education and migration on overall TFRs, and the number of births in the EU28 countries estimated under different projection scenarios. The scenarios differ in terms of the fertility modelling approaches and the migration assumptions they use to illustrate the compositional effects, and in the sensitivity of the models to the assumed migration volumes. We conclude with reflections on the relevance and the limitations of our exercise.

Across social groups, fertility levels, the timing of parenthood, and fertility ideals can vary substantially (e.g., Berrington and Pattaro 2014). Among the socio-economic variables that influence fertility, the educational attainment of women stands out as having a strong and consistent negative relationship with fertility (Rindfuss and Bumpass 1980; Jeejebhoy 1995; Kravdal 2002; Bongaarts 2003; Lutz and Skirbekk 2014). Compared to their less educated counterparts, educated women tend to have fewer children later in life, and are more likely to be childless. Educational expansion is a major driver of the postponement of parenthood, because in modern, post-industrial societies, schooling extends well into reproductive ages (Neels et al. 2017; Ní Bhrolcháin and Beaujouan 2012), and family formation is often normatively conditioned on the completion of education (Blossfeld and Huinink 1991; Mortimer et al. 2005). Analyses of series of completed cohort fertility rates have shown that education differentials tend to decline as the overall fertility rate decreases, and that they tend to be more pronounced among older than younger cohorts. Such differentials are smaller in institutional settings that mitigate the high opportunity costs of childbearing with policies that promote work-family balance, gender equity, and affordable childcare (Kravdal and Rindfuss 2008; Wood et al. 2014). Cohort differentials in fertility by education have diminished in some Western and Northern European countries (van Bavel 2014; Jalovaara et al. 2019), but strong gradients persist, particularly in Eastern Europe (Basten et al. 2014; Wood et al. 2014; Brzozowska 2016).

International migration is another important factor underlying the fertility differences in low-fertility countries. It has become particularly important in the EU members states that host sizeable foreign-born populations, and that are seeing sustained immigration from third countries. In these states, women of reproductive age have become increasingly diverse in terms of their ethnicity, region of origin, religion, and other relevant ethno-cultural characteristics. This diversity could have lasting long-term effects on demographic trends and population composition (Coleman 2006). Across the EU countries, growing shares of births are to mothers of immigrant origin. In some EU member states (MS), the percentage of live births to immigrant mothers has exceeded 25–30% over the past decade.² While births to foreign-born mothers have increased, the net effect on overall TFR remains limited, though not negligible (Sobotka 2008).

As the heterogeneity of European populations is growing, immigrants and their descendants could shape demographic trends to a greater extent in the future than they do today, assuming the current patterns of childbearing and of immigration persist. While immigrants tend to have higher fertility than the native-born population, fertility rates vary greatly among immigrant women depending on their country of origin (Andersson 2004; Dubuc 2012; Mussino and Strozza 2012; Sobotka 2008). In general, women from low-fertility countries tend to have low fertility in the host country, and often have even lower fertility than native-born women; whereas women from high-fertility countries tend to have high fertility in the host country (Mussino and Strozza 2012; Persson and Hoem 2014; Kulu et al. 2017). However, the country of origin may be a less important explanatory variable than the selection of immigrants in terms of their individual characteristics (Andersson 2004; Kulu 2005). Women’s educational attainment, as well as their reasons for migration, likely influence their fertility.

A growing number of studies have highlighted that the country of origin, the age at arrival, and the duration of stay in the host country affect the underlying differences in fertility levels between immigrants. There is strong evidence that the fertility levels of migrants are converging with those of the native-born population. However, for some specific immigrant groups within specific country contexts, the fertility levels of the descendants remain well above the fertility levels of the native-born population (Dubuc 2012; Kulu et al. 2019). Convergence towards mainstream fertility levels is observed across different generations of immigrants, with first-generation immigrants (G1) having the highest fertility rates, those who arrived as children (G1.5) having lower fertility rates, and the descendants of immigrants (G2) having fertility rates that are the most similar to those of native-born women (Kulu et al. 2017; Krapf and Wolf 2016). This pattern of intergenerational convergence has been interpreted through the lens of the adaptation hypothesis, which assumes that the context of the host country greatly influences an individual’s fertility desires, plans, and behaviours. In contrast, the socialisation hypothesis posits that each individual’s preferences are relatively stable over his/her lifetime, and depend on the socialisation context. Thus, according to this hypothesis, an immigrant’s age at arrival, and whether s/he was socialised in a high- or a low-fertility context, will influence his/her observed fertility. Dubuc (2012) found that the convergence, which is prominent for populations originating from high-fertility countries, reflects in part decreasing fertility in sending countries, and in part intergenerational adaptation to the UK mainstream.

The adaptation hypothesis argues that immigrants’ intentions and behaviours change in response to the host country context. Looking at Sub-Saharan African immigrants in France, Afulani and Asunka (2015) found more evidence for the adaptation hypothesis than for the socialisation hypothesis. They also showed that the adaptation hypothesis accounts for the immigrants’ actual fertility, rather than just their fertility ideals.

Fertility is closely intertwined with the migration process. The disruption hypothesis posits that childbearing is likely to be depressed around the time of migration, or may not have started prior to immigration. Consequently, fertility often peaks in the first years after arrival in the host country, resulting in a tempo effect that influences period measurements of immigrant fertility (Toulemon 2004; Mussino and Strozza 2012; Robarts et al. 2015). Persson and Hoem (2014) found for Sweden that the disruption hypothesis is particularly relevant for non-EU immigrants from low-income countries, while non-EU immigrants from high-income countries tend to delay childbearing until several years after migration. Robarts et al. (2015) confirmed both findings for UK immigrants, and showed that (a) the fertility rates of immigrant women from low-income countries peak within four years after immigration; and that (b) women from high-income countries tend to delay childbearing and have lower fertility rates.

The hypothesis of interrelated life events explains the high fertility upon arrival in the host country in the context of the family formation process. According to this hypothesis, immigrants who arrive on family reunification grounds exhibit high fertility in the years immediately after arrival, because for them, childbearing is directly linked to immigration (Castro Martin and Rosero-Bixby 2011; Mussino and Strozza 2012; Robarts et al. 2015). In contrast, women who arrive in the country to work or to study are expected to delay childbearing and have lower fertility. A decision to have a child shortly after arriving in the host country can also be linked to poor employment prospects (Kulu and Milewski 2007). The interrelation between the opportunities in different life domains influences a couple’s fertility decisions.

As immigrants arriving on various grounds can also differ in their educational characteristics, education becomes an important confounding variable to account for when examining fertility differentials. Dubuc (2017) found that educational differences between successive generations of immigrant women and their daughters indeed explained in part the convergence in fertility towards the mainstream.

Thus, making assumptions about the future fertility of immigrant women is challenging, because we would have to account for (a) ongoing fertility transitions in developing countries and (b) educational expansion, and (c) make assumptions about the composition of future migration flows by country of origin. Explicit and implicit migration policies can affect the future composition of immigrants to the EU in terms of their grounds for immigration, their countries of origin (e.g., restrictions on family reunification that favour already established diaspora), their educational levels (e.g., selective policies favouring highly educated immigrants with skills needed in the future labour market), or their gender (e.g., the demand for care workers in ageing societies). Due to selection, immigrants may exhibit very different childbearing patterns than those of the mainstream in the country of origin, as they may be arriving from high-/low-fertility regions, and they could be selected in terms of their educational characteristics.

To conclude, existing evidence of fertility differentials between immigrants and the native-born population, as well as across various immigrant groups, highlights the importance of accounting for three main variables associated with such differentials: the country of origin, the age at arrival in the host country, and the duration of stay. To illustrate the effect of modelling with different sets of differentials – while easing the interpretation of the results – we keep the composition of immigrants,³ as well as fertility differentials among the migrant groups, constant for the whole projection period. As it is well-documented that enrolment in education has a postponement effect on fertility, we argue in the next section that enrolment in education should be accounted for alongside educational gradients in fertility.

The resulting TFRs for the EU28 in the projection scenarios are shown in Figure 2. The increases in the TFRs found for all scenarios result from the general trends set in the assumptions of the reference scenario of Lutz et al. 2018, for more details, see also Basten et al. 2014). The baseline EDU scenario (yellow line) assumes that the TFR increases from 1.65 in 2015–19 to 1.85 in 2050. The differences between the EDU scenario and the additional scenarios reflect the impact of accounting for heterogeneity in fertility behaviours with respect to education and migration, and the variations in the composition of women of reproductive age. If we consider all differentials (by educational attainment, for students and for immigrants), we arrive at lower TFRs that peak at 1.79 children per woman. This result is driven by educational expansion, as illustrated by the EDU + STUDENT scenario that projects a TFR of 1.71 in 2040–59. The projected TFRs in the EDU + STUDENT scenario are lower because in this scenario, the shares of young women enrolled in higher education who have very low fertility while enrolled are increasing. Once we differentiate between the native-born women and various groups of immigrant women (EDU + STUDENT + IMMIG scenario), the overall TFR increases, but to a lower level than in the EDU scenario.

The gap in the fertility levels of EU-born women and foreign-born women is caused by both the educational composition and the migrant characteristics of foreign-born women (Figure 3). In the EDU + STUDENT scenario, in which we do not account for migrant characteristics, foreign-born women have 0.14 to 0.19 points higher TFRs than the TFRs of higher educated EU-born women. In the EDU-STUDENT scenario, the TFRs depend only on the educational composition and the share of women in education.

Once we account for fertility differences across various groups of foreign-born women, the estimated TFR of the foreign-born women increases to 2.5–2.6. The estimated TFR of the EU-born women is about 1.65²¹ (EDU + STUDENT + IMMIG scenario in Figure 3).

The impact of the fertility patterns of foreign-born women on overall fertility depends on the volume of migration and the stock of foreign-born women of fertile age in the baseline population. If both the fertility differences between native-born and foreign-born women and the current migration flows persist at recent levels, the overall TFRs in the EDU scenario and in the EDU + STUDENT + IMMIG scenarios turn out to be very similar. The net effect of immigrant fertility on the overall TFR increases from 0.12 points in 2015–19 to 0.17 points in 2055–59 (Table 2). At current fertility levels, a two-fold increase in immigration from non-EU countries results in a rapidly increasing overall TFR peaking at 1.9 in 2040–44. In the high migration scenario, the net effect of immigrant fertility rises to 0.25 in 2055–59 (Table 2). This finding reflects the rapidly increasing population of immigrant women in the EU population, which is projected to rise from an estimated 10.4 million in 2015 to 38.1 million in 2055 under high migration assumptions, but to just 19.3 million under baseline migration assumptions (Table 3). The share of immigrant women of reproductive age is projected to increase from 10.5% in 2015 to about 30% in 2060 in the high migration scenario, but to only 20% in 2060 in the scenarios with baseline migration (Table 3).

The compositional effects of increased immigration on the overall projected TFRs are more pronounced in countries with larger stocks and projected flows of immigrants from third countries, some of which are listed in Table 2. At current differentials, the net effects of immigrant fertility are projected to be the most pronounced in Sweden, with the projected share of foreign-born women of reproductive age expected to increase from 23% in 2015 to 30% in the baseline migration scenario, and to 39% in the high migration scenario.

In the scenarios with baseline migration, the projected number of births declines from an estimated 26.0–26.5 million in 2015–19 to 23.1–25.7 in 2055–59 (Figure 4). These decreases occur despite the assumed increases in the TFRs (Figure 2) because the number of women of reproductive age declines from 112 million in 2015 to 97 million in 2055 (Table 3), even at sustained recent past immigration volumes. In the doubled migration scenario, the projected population of women of reproductive age increases to 126 million in 2055. In combination with current fertility differentials, this results in a steadily increasing projected number of births in all of the high migration scenarios (Figure 4). When we compare the EDU and the EDU high migration scenarios, we see that a doubling of migration results in 2% more births in 2015–19, with a gradual upward trend to 28% more births in 2055–59. In the EDU + STUD + IMMIG scenarios, larger increases in immigration result in 5% more births as early as in 2015, and 41% more births in 2055 (34.9 million vs. 24.8 million, see also Table 4).

In the EDU scenarios, the number of births is boosted by additional births to students who are assumed to have the same fertility rates as their peers with completed education. The difference in the projected number of births in the EDU and EDU + STUDENT scenarios gradually increases from 2% to 10% over the projection period. The EDU + STUDENT scenario assumes that there are no fertility differentials between native-born and foreign-born women with the same educational levels. If this is indeed the case, and the immigration flows continue to be of magnitudes similar to those in the recent past, the projected number of births will fall below 24 million by 2060.

When we look at the overall TFRS and the projected births in the EDU and EDU + STUDENT + IMMI scenarios, it may appear that modelling fertility with a broader set of differentials does not affect the projection results enough to compensate for the additional complexity in the estimations that need to be performed to arrive at a set of fertility parameters that capture an increased number of dimensions. However, this is the case only if we assume that there is no change in immigration among third-country nationals. A set of scenarios with doubled immigration (dotted lines in Figure 4) show sizeable differences in the projected number of births depending on how the fertility differentials are modelled. Similarly, country-specific results for selected EU countries that are major destination countries of immigrants presented in Table 4 show, with the exception of Germany, a decline in the number of projected births in the EDU + STUDENT + IMMIG scenario with baseline migration, but a sizeable increase in the scenario with doubled migration and at constant fertility differentials. In the case of Sweden, the high migration EDU + STUDENT + IMMIG scenario results in a doubling of the number of births in 2055–59 compared to in 2015–2019. If differences in fertility between the EU-born and the native-born mothers are not taken into account, and only educational differentials are included (EDU + STUDENT), the overall number of births is lower and the surplus in the EDU + STUDENT high migration scenario is smaller than in the full model.

This sensitivity analysis shows that the simple modelling of fertility in the EDU scenario coincidentally captures the effect of the higher fertility rates of immigrant women through the higher fertility rates of women in education at younger reproductive ages. The majority of immigrants arrive as young adults and have elevated fertility in the years immediately after migration, while a sizeable proportion of native-born women are still in education and have low fertility by age 25. Thus, the surplus births in the EDU scenario arise because women enrolled in education are not differentiated from women with completed education. This could, however, change in the future if, for instance, future immigrants arrive mainly as students, or are educated at the same or higher levels than native-born women.

Finally, one of the most interesting and relevant results of our analysis with respect to the increasing diversity of European population is the estimated composition of births by the nativity status of the mother. According to our estimates, if recent migration volumes and fertility differentials persist, this share may increase to 28% by 2055–59 in the EU28 (Table 5). Naturally, the share is largest (39% in 2055–59) in the high migration scenario – nearly double the share in the EDU scenario that includes only educational attainment differentials. Already in the first projection step, we can see a difference of six percentage points compared to the baseline model (EDU) and the full model (EDU + STUDENT + IMMIG) with differentials by region of birth, duration of stay, and generation.

Our results illustrate that in the context of educational expansion and increasing population diversity due to immigration, population projections for the EU countries should account for several sources of population heterogeneity. We have shown that how fertility differentials are modelled affects projection results, especially when looking at projected births. Moreover, we have demonstrated that changes in the composition of women of reproductive age in terms of their educational attainment and migrant status can affect both overall TFRs and projected births.

When studying the impact of education on fertility, it is important to capture the postponement effect that is largely driven by educational expansion. This issue becomes particularly relevant if we assume that post-secondary education will expand in the future, because this expansion will translate into increasing shares of women staying in education until later reproductive ages. We show that the conventional modelling that does not differentiate between the fertility of students and women with completed education results in higher estimated TFRs. Once we account for the lower fertility rates of women enrolled in higher education, the resulting TFRs decline at current fertility differentials. If education differentials were to narrow, or if the negative relationship between fertility levels and educational attainment were to even reverse, the resulting TFRs and the projected births would also be affected. But in these projections, we do not model these hypothetical situations.

A comparison of the EDU scenario and our full model shows that the TFRs and the projected births are similar under baseline migration assumptions. This holds at the current fertility differentials and at the current composition of immigrant women.

The results for the projected births and TFRs are sensitive to migration assumptions. Our high migration scenario shows steadily increasing numbers of projected births, whereas the scenarios with baseline migration based on recent trends project declining births. The net effect of immigrant fertility on the overall TFR in the EU28 is projected to increase from 0.12 in 2015–2019 to 0.17 in 2055–59 in the scenario with baseline migration, and to 0.25 in 2055–59 in the scenario with doubled migration. The net effects and the projected increases in births in countries with larger foreign-born population shares are more pronounced under the assumption of sustained high migration.

We have held the composition of immigration flows constant in order to facilitate the assessment of the impact of fertility modelling on the projection results. Naturally, changing the composition of the flows in terms of the countries of origin is another source of heterogeneity that can impact the projection results. The investigation of the impact of such an exercise is beyond the scope of this paper, but it is feasible using experimental what-if scenarios. If more future immigrants come from high-fertility countries, it is possible that the net effect of immigrant fertility will be more pronounced even if the size of the flow does not change. Fertility is expected to decline to close to the replacement level in some of the major sending countries, and this trend could lead to less pronounced differentials between the native-born and immigrants in the future.

The pace of educational expansion in the countries of origin and the selection of migrants may also change in future. In addition, if more female migrants from third countries start immigrating to study or to seek job opportunities rather than for family reasons, we may see immigrant fertility rates converge more quickly to native-bon levels, and narrower fertility differentials between native-born and immigrant women. A hypothetical situation with no differentials between native-born and immigrant women with same educational attainment is illustrated by our EDU + STUDENT scenario.