Between here and there. Immigrant fertility patterns in Germany.

Transcription

1 Between here and there. Immigrant fertility patterns in Germany. Kamila Cygan-Rehm April 5, 2011 Abstract This study analyses fertility of first generation immigrants using individual-level data taken from the German Socio-Economic Panel (SOEP) and country-specific birth rates reported by the United Nations. We apply generalized Poisson regression to account for the underdispersed nature of the data. The findings support the hypothesis that immigrants are socialized by childbearing standards dominant in their home countries. The more the average birth rates in the home and host country differ, the more fertility of immigrants diverges from the level of natives at destination. We confirm that later migration increases fertility differentials between immigrants and natives. In addition, the extent to which immigrants retain fertility patterns acquired in their home countries depends on the number of years they spent there. Immigrants and natives who completed tertiary education do not follow these general patterns, since their fertility converges. JEL classification: C25, J13, J61 Keywords: migration, fertility, socialization, count data, underdispersion University of Erlangen-Nuremberg, Department of Economics, Lange Gasse 20, Nuremberg, Phone: Helpful comments by Regina T. Riphahn on earlier versions of this paper are gratefully acknowledged. We also thank participants at numerous seminars and conferences for valuable feedback.

2 1 Introduction The list of countries with below-replacement fertility has been getting longer in the recent decades. This has far-reaching consequences for their labour markets and maintenance of social insurance systems. Consequently, the rapid aging of societies became a political key issue and immigration has been recognized as a possible means to slow it down (see e.g. Feichtinger and Steinmann, 1992; Wu and Li, 2003; Alho, 2008). Beyond its direct contribution to the population size in the receiving country, immigration may cause substantial demographic changes if the fertility of immigrants and their descendants is relatively high. Since immigrants contribute remarkably to population dynamics in many contemporary societies (World Bank, 2009), understanding their childbearing behaviour increasingly gains importance to policy makers. Germany could serve as a benchmark case in analysing immigrant fertility because it hosts the largest number of immigrants in Europe. Moreover, over recent decades large migration flows from high-fertility countries coincided with extremely low fertility of natives. Since 1970 the birth rates fell from 2.03 births per woman to 1.37 in 2008 (World Bank, 2009). High fertility of immigrants and their descendants is a widely discussed and a very controversial issue: on the one hand, growing immigrant numbers may be a means against population decline. On the other hand, the insufficient integration, with respect to education, labour market participation, and welfare dependence, increasingly raises concern. The complexity of the childbearing-migration relationship additionally stirs a lively public debate. That migration is not a random process and immigrants tend to be self-selected is well documented. The common literature discusses at least four further hypotheses to explain immigrants reproductive behaviour: socialization, adaptation/assimilation, disruption, and interrelation. Since each of them has received support, and has also been challenged, the exact mechanisms remain unclear. Despite the increasing relevance of the issue, very few empirical studies investigate immigrant fertility in Germany (see e.g. Nauck, 1987; Mayer and Riphahn, 2000; Schmid and Kohls, 2009; Milewski, 2007, 2010). Existing research mostly tests the adaptation hypothesis suggesting that immigrants adjust their childbearing behaviour to the norms of the natives at the new destination. Previous literature indicates heterogeneous fertility patterns across immigrants origins. However, hardly any study adequately examines the socialization hypothesis by testing the role of the home country s birth rates in shaping immigrant fertility. The present paper contributes to the literature in several dimensions. We focus on the lifetime impact of migration on the number of births and 2

3 apply a count data estimation framework. By combining data from two sources we can discern the effect of socialization by home country s childbearing standards and the effect of age at migration on immigrants fertility. Unlike previous studies we examine the reproductive behaviour of all first generation immigrants, not only selected groups. 1 Finally, since distinguishing immigrants and natives by citizenship is inappropriate for the German context, we apply a more adequate classification based on the actual migration experience. 2 We find that both home country s birth rates and age at migration positively correlate with migrants fertility outcomes. In addition, the later in life migration occurs, the more pronounced is the role of home country s childbearing norms. Fertility of immigrants who completed tertiary education is comparable to that of similar natives, regardless of fertility norms dominant in the source country and age at migration. This paper is organized as follows: The next section provides information on the context of German immigration. Section 3 briefly outlines the theoretical framework and reviews previous findings. Section 4 presents our data and section 5 the modelling strategy. Section 6 provides the empirical results. We discuss the findings and conclude in section 7. 2 Immigration and fertility in Germany As of 2008, registered foreigners represented roughly 9 % of the total population in Germany. According to a recently introduced classification almost 19% of the population have a so-called migration background (StBA, 2010). 3 Since East Germany had no significant immigration before re-unification in 1990, the current stock of foreigners results from the long and intense migration experience of West Germany. In the post-war history most immigrants arrived with one of the following streams (see e.g. Schmidt and Zimmermann, 1992): ethnic Germans, traditional guest workers, and humanitarian migrants. Ethnic German repatriates arrived in the aftermath of the World War II, and the dissolution of socialism after They emigrated from former German territories in Central and Eastern Europe, mainly from the former Soviet Union, its successor states, as well as from Romania, Poland, and former Czechoslovakia. Since they were eligible by 1 Nauck (1987) looked at the Turks, Mayer and Riphahn (2000) and Milewski (2007, 2010) at the traditional guest workers from Turkey, Italy, Spain, Greece, and former Yugoslavia, Schmid and Kohls (2009) at Turkish, Greek, Italian, Polish, and former Yugoslavian nationals. 2 See Liebig (2007) for a debate on difficulties associated with using citizenship to define immigrants in Germany. 3 This group comprises persons having migrated to Germany after 1949, foreign nationals born in Germany, as well as German nationals with at least one parent being an immigrant or foreigner. 3

4 law for German citizenship, their migration background is hidden in most official statistics. Traditional guest workers immigrated during the post-war economic boom since the mid 1950s until the early 1970s. Through that time Germany pushed intensive manpower recruitment and signed bilateral treaties with several countries including Italy, Spain, Greece, Turkey, Portugal, and Yugoslavia. Although initially labour migrants residence permit was restricted to one year, they tended to stay longer or even permanently and increasingly brought their family members. Refugees and asylum seekers arrived in the 1990s from the territories under the Yugoslav Wars: Bosnia and Herzegovina, Croatia, Macedonia, Montenegro, Serbia, and Slovenia. The composition of the foreign population currently living in Germany still reflects these major migration streams: the dominant national minorities are Turks, followed by people from former Yugoslavia, Italy, and Poland (StBA, 2010). Despite various geographical roots, the vast majority of immigrants moved from a high to a low fertility context. Table 1 shows the fertility developments in Germany and selected sending countries over the last five decades. [Table 1 about here] The numbers reveal the overall fertility decline starting in the 1970s. Since the late 1980s, total fertility rates (TFR) in all countries save for Turkey have continuously been below the replacement level of 2.1 and nearly converged. Figure 1 presents fertility developments within Germany from 1991 onwards, separately for German and foreign women [Figure 1 about here] While the TFR of German women remains relatively stable at a level of 1.3, the TFR of non-german women successively falls. At the same time, foreign women increasingly contribute to the total number of births. In the observed period the share of births to foreign mothers goes up from 13 to roughly 17 %. 3 Theoretical background and research hypotheses The theoretical framework on immigrant fertility has its roots in various contributions (see e.g. Goldstein and Goldstein, 1981; Hervitz, 1985; Stephen and Bean, 1992; Lee and Pol, 1993; Mulder and Wagner, 1993; Mayer and Riphahn, 2000). The existing literature suggests that the migration event may affect lifetime fertility, the timing of births, or 4

5 both. The most commonly examined hypotheses focus on processes of selection, socialization, adaptation/assimilation, disruption, and interrelation. These five hypotheses are not necessarily mutually exclusive; they are partly complementary, partly contradictory, they may apply to specific lifetime periods and counteract or reinforce one another. We now consider each of them in turn and briefly present previous empirical findings. The socialization hypothesis suggests that immigrants are socialized, i.e. they acquire norms and behavioural patterns regarding childbearing in their home countries and continue to follow them over the life course. When (if ever) the socialization of an individual is completed is not immediately clear. The majority of social scientists define socialization as a life-long process, which is generally divided into two stages: primary and secondary socialization (Mortimer and Simmons, 1978). Primary socialization takes place and is finalized during childhood and adolescence. Only few studies deal with this hypothesis directly (see e.g. Hervitz, 1985; Milewski, 2010), but many raise it indirectly using a different rhetoric (see e.g. Goldberg, 1959, 1960; Stephen and Bean, 1992; Hill and Johnson, 2004). Secondary socialization may occur each time a person encounters a new environment with changed conditions. The migration literature traditionally discusses this mechanism in the context of the post-migratory adaptation. The adaptation, as well as the assimilation hypotheses emphasize the role of the destination country in shaping immigrants reproductive behaviour. The two terms are often used interchangeably because in both cases, sooner or later, immigrant fertility comes to resemble that of the natives. However, the mechanisms behind adaptation and assimilation differ (Hill and Johnson, 2004). The assimilation hypothesis holds that immigrants successively take up the host country s cultural norms regarding family size. Because cultural assimilation takes a fairly long time, we expect it to be more apparent over subsequent generations than within a first generation (Hervitz, 1985; Stephen and Bean, 1992). First generation immigrants may be rather subject to adaptation that starts shortly after migration. Immigrants revise their childbearing preferences as conditions regarding wages, prices, employment, and educational opportunities change. The convergence to the native fertility may be achieved after some years of stay (see e.g. Kahn, 1988; Andersson, 2004) or more precisely with an increasing number of fertile years spent in the host country (Mayer and Riphahn, 2000). Clearly, age at migration determines the duration of exposure to fertility patterns exhibited by the natives at destination. Consequently, previous research interprets the positive effect of age at migration on fertility as a successive adaptation (see e.g. Mayer and Riphahn, 2000; Bleakley and Chin, 2010). However, age at migration outlines also the duration of the socialization process in the country of origin and may positively correlate with fertility for this reason, instead. 5

6 Thus, the exact mechanism behind the pure effect of age at migration on immigrants fertility is ambiguous. Immigrants selection on characteristics additionally tangles the debate on whether fertility of migrants is more similar to fertility of stayers in their country of origin or to residents at the new destination. The process that selects individuals into migration is not random and may result from observable and unobservable attributes such as age, education, employment, motivation, career and family orientation. Since similar characteristics affect childbearing intentions, immigrant fertility may, even before the move, more closely resemble fertility of individuals at destination than of those at home (see e.g. Macisco et al., 1970; Goldstein and Goldstein, 1981; Lee and Pol, 1993; Chattopadhyay et al., 2006). Studies on fertility of international migrants discuss broadly the selection hypothesis, but rarely adequately test it due to lack of bi-national data that allows for comparisons between migrants and their home country s counterparts (Carter, 2000). However, nearly all studies show that dissimilar characteristics play a crucial role in explaining observed fertility differentials between immigrants and natives at destination. In addition, two mechanisms - disruption and interrelation - may lead to a different timing of births shortly before, during, and immediately after migration. The disruption hypothesis suggests that immigrants interrupt childbearing because of inconvenient circumstances associated with the migration process such as separation of spouses or insecurity related to remarkable changes in daily-life conditions. After a temporary drop immigrants may exhibit elevated birth rates leading to final catch up for the initial births lost (see e.g. Goldstein and Goldstein, 1981; Stephen and Bean, 1992). By contrast, the interrelation hypothesis holds that migration and childbearing are interdependent demographic events if the move itself is motivated by marriage, family formation or reunion (see e.g. Mulder and Wagner, 1993; Andersson, 2004; Kulu, 2005). This study analyses the role of age at migration and home country s birth rates in determining lifetime fertility of first generation immigrants to Germany. Based on the earlier discussion we propose four hypotheses: H1: The observed differences in fertility levels between immigrant and native women may be mostly traced back to different characteristics (characteristics effect). H2: Age at migration is a source of considerable fertility differentials between immigrants (age at migration effect). H3: Immigrants lifetime fertility reflects fertility rates dominant in their home countries (socialization effect). H4: The extent to which immigrants preserve fertility patterns of the home country 6

7 depends on number of years they spent there (interaction between socialization and age at migration effects). While we are not aware of any past studies simultaneously examining the effects of socialization and age at migration, the related empirical literature provides useful insides into immigrant fertility patterns in Germany. Nauck (1987) shows that the fertility of Turkish women converges to native levels with increasing duration of stay. Mayer and Riphahn (2000) suggest adaptation for traditional guest workers and show different fertility patterns by country of origin. In her studies Milewski (2007, 2010) also explores reproductive behaviour of traditional guest workers, but in an intergenerational context. Her results indicate assimilation of immigrant descendants to native levels. Schmid and Kohls (2009) confirm heterogeneous patterns in the downward fertility movement for migrants of Turkish, Greek, Italian, Polish, and former Yugoslavian origin. 4 Data, sample and descriptive statistics Similar to previous studies on immigrant fertility in Germany we use individual-level data from the German Socio-Eeconomic Panel (SOEP). The SOEP is a representative longitudinal study of private households, conducted annually since 1984 and the largest dataset providing retrospective information on both, births and migration. Our analysis uses all survey sub-samples. 4 Mayer and Riphahn (2000) show that pure cross-sectional data allow to disentangle cohort and age at migration effects. To increase the spread of included cohorts we pool observations taken from three SOEP waves: 1991, 1999, and 2007 and cluster standard errors by individuals. 5 Since the focus of the study is on the completed fertility, we restrict attention to females aged 45 and above. Additionally, we exclude women who were older than 45 at arrival because their reproductive phase was completed before migration. Based on the respondent s citizenship and past migration experience we construct two mutually exclusive sub-samples, natives and first generation immigrants. To obtain a homogeneous native sample we consider only German citizens who are born in Germany 6 and reside in West Germany 7. The immigrant sample includes all foreign 4 For detailed information on the structure and sampling of the SOEP see e.g. Haisken-DeNew and Frick (2005). 5 We observe 48% of women in our final sample once, 27% twice and 25% three times. 6 We exclude respondents born in Germany but of foreign citizenship who are second generation immigrants. Since parents migration background was not available for the majority of respondents (72%), our native sample includes second generation migrants who took up German citizenship. 7 Fertility and the socio-demographic composition of the East and West German population differ significantly. Moreover, the East Germany had no significant immigration prior to 1991 and even today 7

8 born respondents with immigration experience regardless of their current citizenship. In contrast to the common distinction along citizenship lines, our approach enables us to consider two important immigrant groups: ethnic Germans and naturalized foreigners. Despite their current citizen status they personally experienced migration and we expect them to follow similar fertility patterns as other immigrants. We perform additional test for differences between the immigrants of non-german and German citizenship. Data set limitations do not allow us to distinguish between ethnic Germans and naturalized foreigners in the citizen sub-sample. After these adjustments and omitting observations with missing values on core variables, our sample consists of 7,434 native and 1,163 immigrant females. Almost one third of the sampled immigrants are German citizens. Table 2 provides descriptive statistics of the dependent and relevant explanatory variables, separately for the native and immigrant sub-samples. [Table 2 about here] We observe considerable differences in fertility between the sub-samples: the average native fertility with 1.90 births per woman is far below the immigrant mean of 2.60 births. Previous research and theoretical considerations guide our selection of background variables. In particular, we consider four groups of individual characteristics known to determine childbearing choices. First, we control for woman s birth year to account for cohort effects on fertility level. Immigrants are on average six and one half years younger. Second, we approximate woman s opportunity costs of an additional child by her educational attainment. 8 Because education is as a powerful predictor of fertility, we consider both, years of schooling and a threshold variable for the highest completed degree distinguishing the elementary, secondary, and tertiary level (Hill and Johnson, 2004). Immigrants receive on average almost two years less education and have lower secondary and tertiary graduation rates. Third, we consider woman s family orientation, in terms of her unobserved attitudes towards traditional gender roles and family structures, desired family size, contraception and birth control. As proxies we include observable factors such as an indicator of whether she was ever married, her age at first marriage, the number of her siblings, 90 % of foreign nationals live in the western part of the country. 8 Direct measures of opportunity costs such as woman s income and employment are inappropriate for the issue of fertility because of their endogeneity. For the same reason we do not consider woman s current self-reported German language skills, which is the only available measure of language fluency in SOEP. 8

9 and her religious affiliation. The literature on intergenerational fertility transition suggests that women raised in larger families are more likely to bear more children than women from smaller families. Immigrants have on average notably more siblings, while the differences in marriage behaviour are moderate. Since previous research strongly emphasizes the role of religion in determining fertility, we control for woman s religious affiliation and her religiosity represented by the frequency of attendance to church or other religious events. The religious affiliation in our sub-samples differs substantially. Most notably, while jointly almost 88 % of native women are Christians (either catholic or protestant) and almost none of them is Muslim, the religious affiliation of immigrants is much more diverse. Furthermore, immigrants tend to attend church or other religious events more frequently. 9 Finally, we include individual migration background to determine the effect of being an immigrant on lifetime fertility. Since the immigrant population is heterogeneous, in a separate model specification we distinguish between immigrants with and without German citizenship. Small sample sizes of immigrants from several countries preclude us from using controls for country of origin. We group countries together and define six immigrant sub-populations instead. Most sampled immigrants originate from Southern European countries of guest workers recruitment: Italy, Spain, Greece and Portugal. Women of Turkish origin alone account for 21 % of the immigrant sample. A notable share of 27 % arrived from former Yugoslavian territories. The remaining immigrants are mostly of Central and Eastern European descent. Immigrants from Russia, Kazakhstan, Poland, and other Eastern European countries jointly amount to about 40 % of the sub-sample. We include as further covariates age at arrival and a measure for the extent to which fertility patterns in immigrant s country of origin and Germany differ. We construct this proxy variable using country-specific total fertility rates (TFRs) reported by United Nations. We assign to each immigrant woman both, the TFR in her home country and the corresponding German TFR as of the migration year. Next, we calculate the absolute difference between the two. 10 Sampled immigrants are on average 28 years old at arrival and move from countries where the TFR is by 1.13 births higher than the German one. Almost 85 % of sampled immigrants move from a high to a low fertility context. Table 3 shows the average number of births for different categories of the two variables 9 Since religious affiliation has been asked neither in 1991, nor in 1999, we use the first religious affiliation a woman ever reported to SOEP. Alternative specifications for this variable including indicators for whether a woman ever belonged to a particular religious community provide nearly identical results. 10 This variable originally ranged from to 5.88 births per woman. We excluded 1% of outlying observations for which the value exceeded

10 of interest. Both, age at migration (0.21) and the difference in national TFRs between home and host country (0.36) significantly correlate with the dependent variable in the immigrant sample. [Table 3 about here] Two patterns are apparent: first, the younger a woman is at migration, the fewer children she bears. Whereas women arriving at ages give on average 3.09 births, those who migrate earlier in life exhibit fertility much closer to the native level of Second, the greater the difference in national TFRs between the home and host country at arrival, the higher immigrant s fertility. To estimate the extent to which the higher immigrant fertility is attributable to selection for characteristics, age at migration, and socialization we apply a multivariate approach described next. 5 Estimation strategy Given that the dependent variable, the total number of births, is a non-negative integer we apply a count data approach based on maximum likelihood (ML). The natural starting point for analyses of counts is the standard Poisson model including tests of the underlying equidispersion assumption E[y i X i ] = V [y i X i ]. As in other fertility studies (see e.g. Winkelmann and Zimmermann, 1994; Wang and Famoye, 1997; Mayer and Riphahn, 2000), regression-based tests (see e.g. Cameron and Trivedi, 2005) indicate that the conditional mean exceeds the conditional variance in our data. This underdispersion may lead to overestimated standard errors. The relevant literature proposes several more flexible approaches to generalize the Poisson model and to obtain reliable standard errors from underdispersed data via ML (see e.g. Consul, 1989; Famoye, 1993; Winkelmann and Zimmermann, 1994; Wang and Famoye, 1997). The most common solution is to relax the equidispersion condition by assuming the variance to be proportional to the mean. Following Joe and Zhu (2005) we apply a generalized Poisson regression (GPR) based on the first two conditional moments given respectively by E[y i X i ] = exp(x iβ) 1 δ V [y i X i ] = exp(x iβ) (1 δ) 3 = 1 (1 δ) 2 E[y i X i ] (2) (1) 10

11 where y i stands for the dependent count variable, X i for the vector of covariates, and δ for an additional dispersion parameter: when δ = 0 the GPR reduces to the standard Poisson model. For δ < 0 the GPR represents count data with underdispersion. We use the STATA program gpoisson by Hardin et al. (2007), which estimates δ simultaneously with the coefficients β and provides a Likelihood-ratio test of the hypothesis of equidispersion, i.e. H 0 : δ = 0. We estimate several models that differ with respect to X to test our hypotheses. We begin with a simple model that controls only for cohort effects. The results give preliminary evidence for gross differences in fertility between immigrants and natives. Next, to account for the selectivity of immigrants (H1) we additionally introduce woman s educational attainment, an indicator of whether she was ever married, age at first marriage, number of siblings, and religious affiliation. We expect that the observed fertility differentials between immigrants and the natives diminish once we control for the socio-demographic characteristics. In the next step we test the positive effect of age at migration on fertility outcomes reported in the previous literature (H2). Simultaneously, we examine to what extent the fertility divergence results from different childbearing norms, which immigrants become acquainted with prior to migration (H3). We accommodate this previously ignored effect by our proxy for the observable difference in the birth rates between the immigrant s country home and Germany at the time of move. A positive effect would indicate the socialization to be at work, i.e., the more the TFR in immigrant s home country exceeds the German one, the higher immigrant s final fertility. So far our modelling strategy has focused on assuming constant effects on lifetime fertility. However, the socialization effect on fertility may be the more persistent, the more years immigrants spent in the home country (H4). Therefore, the final model allows the effect of difference in TFRs between immigrant s home and host country to vary with age at migration. The next section discusses the estimation results. 6 Results Table 4 presents the estimation results for four main model specifications. 11 The negative and significant parameter δ confirms the underdispersion in our data. 11 Standard Poisson regressions yield similar results in all model specifications. Also, OLS regressions provide identical signs of the estimated coefficients throughout. We tested the robustness of our results to cross-sectional weighting and drop of duplicate observations. These estimates generally indicate greater coefficients for the difference in TFRs and the interaction term, and smaller coefficients for the immigrant dummy and age at migration. However, plots show that the findings do not change qualitatively. Detailed results are available from the author upon request. 11

12 [Table 4 about here] Our regression aims to measure the effect of being an immigrant on lifetime fertility based on a comparison with the natives at destination. 12 Central are the estimation results for variables representing migrant-specific characteristics shown in the upper part of Table 4. We begin with a simple model that presents fertility differences between immigrants and natives adjusted for birth cohort only (column GPR1). As expected, the coefficient of the immigrant indicator is positive and significantly different from zero (at the 1% level). Since we may approximately interpret single coefficients as semielasticities, we conclude that in general the final fertility of immigrants is roughly by 39% ((exp(0, 331) 1) 100) higher. After controlling for a wide range of observed socio-demographic characteristics the gross fertility differentials between immigrants and the natives diminish (column GPR2). Thus, our estimation results confirm the selectivity of immigrants on characteristics as stated in hypothesis H1. Still, immigrant women give on average significantly more births than native women with identical characteristics. Almost all of the control variables are important predictors of fertility outcomes, and they affect fertility in the expected direction. The coefficients of the birth cohort dummies reveal the overall declining fertility in younger cohorts, which is commonly explained by a more effective use of birth control. The three cohort indicators are jointly significant in each of the four specifications. Not surprisingly, the coefficient for years of schooling confirms the negative effect of human capital on childbearing. Clearly, we may interpret the effects of secondary and tertiary education only jointly with the continuous effect of years of schooling. The negative coefficients for the educational threshold variables are in accordance with previous literature suggesting higher transition rates to a second and higher order birth for women who completed at least secondary education. The results for the variables describing family orientation and religious affiliation are also consistent with earlier findings on fertility determinants. Most notably, being ever married and early marriage are associated with higher fertility outcomes. Muslims exhibit substantially higher fertility than women of any other religion. We observe that the estimated coefficients for the control variables do not change notably in later model specifications. To evaluate hypotheses H2 and H3 we next examine the effect of age at migration and the effect of the proxy for the difference in childbearing norms between the home and host country (column GPR3). For both included variables we could reject the 12 Since one could argue that the decision to remain childless may be driven by different mechanisms, we repeated all estimations only for women who gave at least one birth. These results confirmed the general patterns and thus are not reported. 12

13 hypothesis that a higher order polynomial improves the goodness of fit at high levels of significance. As expected, both age at migration and difference in national TFRs are positively related to fertility outcomes. However, since both analysed covariates represent interaction terms between an immigrant-dummy and a corresponding variable, the interpretation of estimated coefficients and their significance is not straightforward. In a non-linear regression the signs and magnitudes of estimated effects may change for different values of all included covariates. Consequently, the statistical inference cannot build upon simple t-tests of individual coefficients for migrant-specific variables (Ai and Norton, 2003). Instead of focussing on the results in Table 4, we draw on Greene (2010) and perform a graphical analysis of the estimated effects. Figure 2 plots the relative fertility differentials between immigrants and natives as a function of age at migration. [Figure 2 about here] The effect is calculated for an average immigrant arriving from a country where the TFR exceeds the German one by 1.13 births. For both natives and immigrants we fix all remaining control variables at the means of the pooled native-immigrant sample. The dashed lines represent pointwise 95% confidence intervals obtained by 1,000 bootstrap iterations. As stated in H2 age at migration is an important determinant of fertility outcomes. Immigrants arriving at age 28 give roughly 9.7% more births than comparable natives. An additional year spent in the home country increases the outcome to 11%. Immigrants who postpone the move until age 38 overtake the native level by at least 23.3%. The depicted effect is not statistically significant at the 5% level over the entire age range. We find that the significant fertility deficit of immigrants arriving prior to age 11 is a result of specification. A model including additional interactions with an indicator for childhood migration provides an insignificant effect for this immigrant group. 13 Based on these findings, we reject hypothesis H2 for women who migrate as children or as young adults (conditional on an average difference in country-specific TFR of 1.13). Figure 3 represents a test for the socialization hypothesis H3. [Figure 3 about here] The solid line depicts immigrants excess fertility compared to natives when plotted over the range of the variable difference in national TFRs. We investigate the effect for 13 We define a childhood immigrant as a woman who was under age 15 at arrival. Detailed results are available from the authors upon request. 13

14 age at migration 28. Similar to our earlier proceeding we keep all remaining variables constant. The shape of the curve confirms the positive effect of the difference in national TFRs on fertility differentials between immigrants and natives. Immigrants from countries where the TFR exceeds the German one by 1.1 births bear on average 9.5% more children than comparable natives. The greater the difference in country-specific TFRs, the more immigrants completed fertility diverges from the native level. An increase in the difference in national TFRs to 1.2 births is related to a growth in fertility excess vs. natives by 0.6 percentage points. However, the estimated effect is statistically insignificant at the 5% level for values up to a difference in TFRs of roughly 0.7 births. About 55% of the women in our immigrant sample arrive from this kind of fertility context. Still, our evidence favours the socialization hypothesis for immigrants arriving (at average age) from fairly higher fertility contexts. To test hypothesis H4 we interact the two previously discussed effects in the final model specification (column GPR4). The estimated coefficient for this interaction term is positive and statistically significant at the 5% level. This result indicates that the extent to which fertility norms acquired in the home country shape immigrants fertility is affected by the number of years spent there. However, this model specification shows that the home country s birth rate by itself has no significant effect on fertility. Estimated coefficients for all immigrant-specific variables are jointly significant at the 0.1% level. We do not wish to stress the numerical results too much and instead present our results graphically in Figures 4 and 5. [Figures 4 and 5 about here] The figures reveal how the effect of the difference in national TFRs varies with the age at migration. We show this effect if measured at ages 22 and 34, which refer to the 1st and 3rd quartile in our immigrant sample. The overall conclusion does not change: the more contrary the childbearing standards in the home and host country, the less immigrant fertility converges to that of natives. However, the slope, cut-point with the reference native level, and significance of the estimated effect plays out differently by the number of years spent at home. The almost flat fertility profile in Figure 4 suggests that women who arrive at age 22 exhibit no significant fertility differences if compared to natives, regardless of the difference in national TFRs between the home country and Germany. This result remains unchanged for immigrants arriving prior to age 26 (not presented). In contrast, the steep curve in Figure 5 indicates that immigrants coming at older ages bear significantly more children even if they arrive from fairly resembling fertility contexts. Clearly, the higher the age at migration, the sharper the effect of the 14

15 difference in national TFRs on fertility. Finally, repeated calculations for the entire age range show no significant excess fertility, regardless of age at migration, if immigrants arrive from childbearing standards that are identical to the German ones. The difference in national TFRs of 0.1 births is already associated with significant fertility divergence for immigrants arriving by the end of their fertile years (starting at age 38). Overall, we take these findings as evidence for hypothesis H4, suggesting that the way in which home countries birth norms determine immigrants fertility depends on the time spent at home. So far we estimated the average effects. However, other factors like unobserved cultural background, migration strategy, or motivation for migration, may be sources of considerable differences among immigrants. To capture systematic cultural differences other than childbearing norms, we consider immigrants countries of origin. Neither separate estimations by country of origin group nor controlling for the country of origin group reveal reliable results: limited variability in the variable difference in national TFRs within the single groups and its high (up to 0.88) and significant correlations with the indicators for country group do not allow us to identify remaining cultural effects on immigrant fertility. Nevertheless, we estimate the coefficients for migrant-specific covariates separately for immigrants with and without German citizenship to account for potential differences resulting in migration strategy and experiences. [Figures 6 and 8 about here] From Figures 6 and 7 we derive that: first, on average fertility differences to natives are more pronounced among immigrants of non-german than among those of German citizenship. These discrepancies are presumably driven by the mechanisms of pre-selection into naturalization and suggest that the results of previous studies, which analyse only immigrants of non-german citizenship, may be upward-biased. Second, while increasing age at migration affects non-citizens more than the citizens, the differences in childbearing standards between the source country and Germany determine fertility of both groups to similar degree. By comparison of Figures 7 and 8 (calculated for age at migration 28 and 34, respectively) we observe that the reinforcing effect of age at migration on persistence of socialization is moderately steeper for immigrants of German citizenship. Still, since the confidence bands overlap, the results indicate that both immigrant subgroups follow qualitatively similar fertility patterns. Finally, we repeat the analysis by educational threshold. As shown in Table 5 the determinants of fertility in general vary across the groups. Most notably, while religious 15

16 affiliation appears to be relatively important for childbearing choices of women with elementary and secondary degrees, it has little effect on higher educated women. The latter are more responsive to changes in individual family background variables like number of siblings, marriage, and age at first marriage. [Table 5 about here] Considering the total effect of four immigrant-specific covariates, we notice several important differences. The results for the lowest education level depicted in Figures 9 to 11 confirm the general patterns obtained on the entire sample with one exception: with increasing age at migration the slope of the effect of difference in national TFRs changes only moderately. Still, the coefficients for this variable and its interaction with age at migration are jointly significant at the 10% level. All four immigrant-specific coefficients are jointly significant at the 1% level. [Figures 9 to 11 about here] The inspection of the estimated effects for the group with secondary degree plotted in Figures 12 to 14 leads to slightly different conclusions. [Figures 12 to 14 about here] While the directions of the effects remain unchanged, increasing age at migration affects the effect of difference in national TFRs stronger than in case of the less educated group. Presumably as a result of small sample size, none of the plots shows statistically significant excess fertility compared to similar natives. We find that the combination of late migration (after age 36) and extremely opposite national TFRs (starting for difference of 2.0) leads to meaningful and statistically significant fertility differentials of at least 30% in this group (not shown). All four immigrant-specific coefficients are jointly significant at the 10% level, and without the immigrant indicator at the 5% level. In contrast, the fertility curves for women who completed tertiary education displayed in Figures 15 to 17 are nearly flat and partly of the opposite directions. [Figures 15 to 17 about here] Fertility of immigrants in this group is not subject to socialization by home country s childbearing norms. Regardless of age at migration and the home country s birth rate, fertility levels of highly educated immigrants and comparable natives converge. The immigrant-specific coefficients estimated for this educational threshold are jointly insignificant in either combination. 16

17 7 Conclusion This study shows several remarkable patterns in childbearing behaviour of first generation immigrants in Germany. First, similar to immigrants in other West European countries, immigrants in Germany give on average significantly more births than natives. We confirm that different socio-demographic characteristics, in particular immigrants lower educational attainment largely explain higher fertility levels. Second, age at migration is an important determinant of fertility differences among immigrants. This finding is consistent with prior evidence suggesting that early migration reduces lifetime fertility. We observe that fertility levels of immigrants who arrived prior to age 26 do not differ significantly from the native level. Since women who are younger at arrival are exposed to the lower fertility norms at the destination for a longer time, previous research interprets this effect as time-dependent adaptive behaviour (see e.g. Mayer and Riphahn, 2000). However, by the same logic we argue that age at migration determines also the duration of the socialization process in the home country. Moreover, our results indicate that fertility differences between immigrants arriving after age 26 and natives are distinctly lower than shown in earlier research. 14 These discrepancies may partly result from different sample restrictions. While previous studies consider only immigrants of non-german citizenship, we additionally include ethnic Germans and naturalized foreigners. Still, the estimated excess fertility compared to natives falls clearly below the previously reported one. 15 This finding refers to the entire range of age at migration and suggests that the previously ignored effect of birth rates prevailing in the country of origin plays an important role in explaining fertility differentials between immigrants and natives. In particular immigrants from fairly opposing fertility standards follow their home country s fertility patterns over their life course. Fertility of immigrants coming from identical childbearing standards coincides with the native level regardless of age at migration. We take these findings as evidence of socialization, i.e. that immigrants acquire norms regarding family size in their country of origin. Finally, our results indicate that the extent to which immigrants preserve their home country s fertility patterns depends on the number of years they lived there. Premigration socialization determines immigrants fertility over the life course the more, the later migration occurred. However, separate estimations by educational threshold suggest that the fertility of highly educated women is not subject to these mechanisms. 14 While our model predicts fertility excess of roughly 12% for women arriving at age 30, Mayer and Riphahn (2000) suggest the difference to be about 51%. 15 Our results indicate excess fertility for foreign citizens of roughly 20%, given that they arrived at age 30. For immigrants being German citizens the difference to the natives is 6%. 17

18 We do not observe significant fertility differences between immigrants and natives if they completed tertiary education, independent of age at migration and home country s birth rates. We conclude that childbearing behaviour of first generation immigrants is affected by both: age at migration and the fertility patterns that immigrants acquire in their home countries. Since over recent decades birth rates in the major source countries of immigrants living in Germany have gradually declined (World Bank, 2009), we may expect that fertility of newly arriving immigrant cohorts will successively approach the low native level. This convergence may be additionally accelerated by an observable trend towards higher education among younger immigrant cohorts (see e.g. Seifert, 1997). After all, educational attainment appears to be a crucial determinant of childbearing choices. 18

19 Appendix Figure 1: Fertility in Germany by woman s citizenship Note: Upper part: TFR by citizenship, TFR of 2.1 is considered to be replacement level. Bottom part: share of births to foreign mothers in total number of births. Source: German Federal Statistical Office. 19

20 Figure 2: Excess fertility relative to natives as function of age at migration Note: Solid line: Relative difference in predicted number of children between immigrants and natives calculated for the mean difference in national TFRs of Remaining covariates are fixed at the means of the pooled native-immigrant sample. Dashed lines: 95% pointwise confidence bands based on bootstrap with 1,000 repetitions. Source: Own calculations based on SOEP, pooled waves: 1991, 1999 and National TFRs from The United Nations. 20

21 Figure 3: Excess fertility relative to natives as function of difference in national TFRs Note: Solid line: Relative difference in predicted number of children between immigrants and natives calculated for the mean age at migration of 28. Remaining covariates are fixed at the means of the pooled native-immigrant sample. Dashed lines: 95% pointwise confidence bands based on bootstrap with 1,000 repetitions. Source: Own calculations based on SOEP, pooled waves: 1991, 1999 and National TFRs from The United Nations. 21

22 Figure 4: Excess fertility relative to natives as function of difference in national TFRs for immigrants arriving at young ages Note: Solid line: Relative difference in predicted number of children between immigrants and natives, calculated for age at migration of 22. Remaining covariates are fixed at the means of the pooled nativeimmigrant sample. Dashed lines: 95% pointwise confidence bands based on bootstrap with 1,000 repetitions. Source: Own calculations based on SOEP, pooled waves: 1991, 1999 and National TFRs from The United Nations. 22

23 Figure 5: Excess fertility relative to natives as function of difference in national TFRs for immigrants arriving at older ages Note: Solid line: Relative difference in predicted number of children between immigrants and natives, calculated for age at migration of 34. Remaining covariates are fixed at the means of the pooled nativeimmigrant sample. Dashed lines: 95% pointwise confidence bands based on bootstrap with 1,000 repetitions. Source: Own calculations based on SOEP, pooled waves: 1991, 1999 and National TFRs from The United Nations. 23

24 Figure 6: Excess fertility relative to natives as function of age at migration by immigrant s citizenship Note: Solid lines: Relative difference in predicted number of children between immigrants and natives, separately for immigrants of German (grey line) and non-german (black line) citizenship. Both effects are calculated for difference in national TFRs of 1.13 and remaining covariates at the means of the pooled native-immigrant sample. Dashed lines: 95% pointwise confidence bands based on bootstrap with 1,000 repetitions. Source: Own calculations based on SOEP, pooled waves: 1991, 1999 and National TFRs from The United Nations. 24

25 Figure 7: Excess fertility relative to natives as function of difference in national TFRs by immigrant s citizenship Note: Solid lines: Relative difference in predicted number of children between immigrants and natives, separately for immigrants of German (grey line) and non-german (black line) citizenship. Both effects are calculated for age at migration of 28 and remaining covariates at the means of the pooled native-immigrant sample. Dashed lines: 95% pointwise confidence bands based on bootstrap with 1,000 repetitions. Source: Own calculations based on SOEP, pooled waves: 1991, 1999 and National TFRs from The United Nations. 25