Education and the Transition to Fatherhood: The Role of Selection Into Union

Abstract

Introduction

A major fertility trend of the past decades in the West has been the postponement of parenthood. Chief explanations of postponement include the expansion of women’s enrollment in advanced education and their increased participation in the labor market. More highly educated women, who are also more likely to be active in the paid labor market, tend to make the transition to parenthood at a later age than their lower-educated peers, even if the former often catch up at later ages (Mills et al. 2011; Sobotka 2004; Sweeney 2002).

The role of men’s education in the transition to parenthood has received much less attention. Focusing on the role of men’s education is important because major changes have taken place in the relative education of men and women. Although men were typically more educated than women in the past, the gender gap in education has reversed in most Western countries. In recent years, the number of highly educated women reaching the reproductive ages is exceeding the number of highly educated men (DiPrete and Buchmann 2006; Van Bavel 2012). This shift has affected educational assortative mating: educational homogamy remains dominant, but the female partner’s education now typically exceeds the male partner’s (hypogamy) in case of differential attainment levels, whereas the reverse (hypergamy) has always been true in the past (Esteve et al. 2012; Grow and Van Bavel 2015; Schwartz and Mare 2005). This reversal has potentially far-reaching consequences for family formation (Van Bavel 2012).

Parenthood implies parental investments and resources from both women and men. The decline of the male breadwinner–female homemaker model has turned obsolete the “separate spheres” argument for focusing only on women’s characteristics. Women’s increased earning potential and activity in the labor market may put pressure on men to be more actively involved in household work and childcare activities (Huinink and Kohli 2014; Martin-Garcia 2009; McDonald 2000; Sweeney 2002). Although time-use studies have shown that men still spend much less time than women on household chores and childcare (Craig and Mullan 2010, 2011), men and women have, on average, become more equally involved in daily parenting activities compared with earlier generations. In the context of the dual-earner family, education plays an important role in shaping gender relations in family formation processes, as well as gender relations in established households (Carlson et al. 2013; Goldscheider et al. 2014; Martin-Garcia 2009; Van Bavel 2012).

The findings about the effect of men’s education on the transition to parenthood have been inconsistent. We argue that this inconsistency exists because earlier studies of men’s education and family formation have typically looked at it either in the context of an established couple or without simultaneously considering the education of the female partner. Studies investigating fertility from the couple’s perspective fall into the first category (see, e.g., Begall 2013; Corijn et al. 1996; Jalovaara and Miettinen 2013; Thomson 1997; Vignoli et al. 2012). A limitation of such studies is that they suffer from selection bias: because only partnered men are analyzed, these studies lose sight of how some men are selected into unions whereas others are not. The second type of studies that have looked at the effect of male characteristics do not suffer from this bias, but at the expense of failing to control for the effect of female characteristics. In the literature about the transition to adulthood, scholars have investigated the effect of education both for men and women; but in order to be able to include singles, the effect of the partner’s education had to be left out of the equation (Billari and Liefbroer 2007; Corijn and Klijzing 2001). Such study design fails to account for strong educational homogamy, and as a result, one cannot tell to what extent the estimated effect of his education really reflects his rather than her education.

In this study, we propose to combine the advantages of both approaches: namely, to include both single and partnered men while taking into account the education of the female partner, if any. We do this by simultaneously modeling union formation and the transition to parenthood as two interrelated processes. We hypothesize that men’s education has a consistent and positive effect on fatherhood rates but that this acts chiefly through the process of union formation: men with higher educational attainment tend to be more attractive on the mating market and therefore exhibit higher rates of union formation. As a result, they also exhibit higher fatherhood rates. If only men with a partner are considered—after the selection effect has played out—the effect of his education on top of the effect of the female partner’s education becomes much more uncertain, which is why estimates from couple-level studies may be inconsistent. Within couples, it is typically her rather than his education that affects fertility outcomes.

To test the selection-into-union hypothesis for men, we fit a multiprocess model to account for the endogeneity of union formation and parenthood. The literature considering the endogenous relations between family events has mostly focused on women’s characteristics (see, e.g., Baizán et al. 2003; Brien et al. 1999). To date, there is a lack of studies focusing on the link between men’s education and the transition to fatherhood and on the relationship between union formation and fatherhood. This study aims to fill that gap. We replicate our multiprocess model in 10 European countries, using data from the Gender and Generations Surveys (GGS), to explore the sensitivity of our hypothesis to different European contexts. Our findings do indeed show consistently positive effects of education on the transition to fatherhood in diverse European contexts for men, but they operate mainly through union formation.

In the following sections, we discuss theoretical insights about the relationship between men’s education and family formation and then review lessons learned from earlier empirical studies. Next, we explain more about our empirical strategy and report on our findings.

Education and Men’s Family Formation

Given its multidimensional nature, education may affect union formation and the transition to parenthood through several mechanisms (Kravdal and Rindfuss 2008; Lappegård and Rønsen 2005; Tesching 2012; Van Bavel 2010). These mechanisms may differ between women and men and will play a different role depending on the social context. Our theoretical framework focuses on the transition to fatherhood within the context of cohabiting or married unions, as is the case for the vast majority of childbirths in Europe and the United States (Cherlin 2010; Manlove et al. 2010; Perelli-Harris et al. 2012). Our framework may be less adapted to analyze the transition to fatherhood among men who are not coresiding with (the mothers of) their children. Such situations occur more often at younger ages and are more common among men with low socioeconomic status and education, particularly in the UK and the United States (Berrington et al. 2005; Rendall et al. 1999; Zhang 2011). Fatherhood outside of unions often leads to a situation where the biological fathers are hardly fathering their children actively (Berrington et al. 2005; Eggebeen and Knoester 2001; Goldscheider and Kaufman 1996). For a recent discussion of relevant theoretical perspectives about fatherhood outside unions, we refer to Carlson et al. (2013).

In the literature about the interplay between union formation and parenthood, two types of economic mechanisms are usually distinguished that relate his and her education with childbearing decisions: the positive income effect and the negative price effect. The income effect accounts for the fact that the more-educated people tend to earn a higher income and are therefore more likely to afford the monetary costs of having (additional) children. The price effect, on the other hand, acts through opportunity costs: highly educated people have high opportunity costs because they have more to lose when they have to devote more time to nonpaid activities, such as childcare and household chores, after becoming a parent (Becker 1991).

In the male breadwinner model, opportunity costs predominate for women, whereas the income effect is more important for men. In this model, the expected educational gradient in union formation and fertility is negative for women and positive for men because the opportunity costs of family formation are larger for college-educated women, whereas the positive income effect predominates for their male peers. Even before graduating, women have more difficulty balancing the role of wife/mother with that of student, so the negative effect of educational enrollment is expected to be stronger for women than for men (Blossfeld and Huinink 1991; Liefbroer and Corijn 1999).

The male breadwinner model has now largely given way to a dual-earner model (Sweeney 2002). Since the last decades of the twentieth century and the first decades of the twenty-first century, gender inequalities at macro and micro levels have changed. In education, the gender gap has reversed. Since approximately the 1990s, female enrollment in college-level education in most of the OECD countries exceeds male enrollment, and women also complete their education more successfully (Vincent-Lancrin 2008). Female labor market participation has increased dramatically, and to a much lesser extent, men have also become more involved in household chores and childcare (England 2010; Oppenheimer 1994; Raley et al. 2012). Women’s earnings in the labor market are increasingly considered an essential part of the family budget (Oppenheimer 1994).

When the dual-earner model prevails, the expected relationship between educational attainment and family formation is different. Women with high earning potential increasingly become more attractive partners on the mating market (DiPrete and Buchmann 2006; Oppenheimer 1988, 1994; Sweeney 2002). For them, the positive income effect may start to prevail, although they may dampen opportunity costs either by outsourcing childcare and household chores (Kravdal 2007) or by sharing household work more equally with the male partner (Sullivan et al. 2014). For men, in turn, the opportunity costs may rise because men are under pressure to increase their engagement in parenting and housework (Huinink and Kohli 2014).

Not only does an individual’s own education matter for family formation, but the partner’s education is also relevant. In a large majority of couples, both partners have the same level of education. However, the reversal of the gender gap in education has changed the situation in couples with a difference in education. Historically, the male partner typically had as much as or more education as the female partner; after the reversal, women typically are equally or more educated than their partners (DiPrete and Buchmann 2006; Esteve et al. 2012; Grow and Van Bavel 2015). This shift is associated with an increased proportion of families where the woman is the main breadwinner, even if a gender pay gap to the disadvantage of women persists (Klesment and Van Bavel 2015). Under these new circumstances, the significance of the educational attainment of the male partner may change. The income of the male partner may become a less crucial selection criterion who themselves have a high income. High-earning women may become more interested in the social fathering skills of potential partners. Men with less education may compensate a limited income potential by exhibiting the will and ability to be involved in household chores and child-rearing tasks. In this way, they may enhance their attractiveness to college-educated women who want both a career and a family (Van Bavel 2012).

To the extent that the dual-earner family becomes the norm, the impact of education on family formation will tend to become more similar for men and women, with opportunity costs of parenthood for men and income effects for women becoming more salient. At the same time, the impact of men’s own education will also depend on the education of the chosen spouse, and vice versa, implying that the effect of education on the transition to parenthood cannot be understood without considering the selection effects involved in union formation.

Beyond selection effects, the interdependencies between union formation and the transition to parenthood complicate the distinction between cause and effect. For men in particular, accounting for the interrelatedness of union formation and parenthood is crucial. Finding a suitable partner is a necessary prerequisite to become a father. After finding a partner, the transition to fatherhood strongly depends on the stability of the union and the characteristics of the spouse. Coresidence typically implies an acceleration of the family formation process. Conversely, looking in the other causal direction, a pregnancy may expedite coresidence (Baizán et al. 2003). The interrelationship between union formation and parenthood is strengthened by the fact that individuals are heterogeneous in factors that may simultaneously affect both kinds of events. Some of these factors are observed, such as education, but many are typically unobserved, such as personality traits and physical characteristics. A proper empirical analysis should account for these interdependencies.

Earlier Empirical Findings

Studies addressing men’s transition to parenthood tend to fall into one of two categories. First, life course research about the transition to adulthood has typically looked at men and women separately, investigating variability in the occurrence, order, and timing of events. Second, studies that focused on fertility from a couple’s perspective typically looked at the influence of male characteristics after controlling for female characteristics.

Studies about the transition to adulthood have consistently shown that school and college enrollment delay union formation and parenthood, for men as well as for women (Blossfeld and Huinink 1991; Corijn and Klijzing 2001). Enrollment delays parenthood more than union formation, and the effect is found to be weaker for men than for women (Corijn and Klijzing 2001; Liefbroer and Corijn 1999; Winkler-Dworak and Toulemon 2007). The effect of educational attainment is much less clear than the enrollment effect. Some studies found that high attainment accelerates men’s union formation and marriage (Corijn and Klijzing 2001; Goldscheider and Waite 1986; Winkler-Dworak and Toulemon 2007). Kalmijn (2011, 2013) showed that in Europe, men with better career prospects and positions in the labor market have higher chances of forming a union and getting married, whereas unmarried cohabitation is related to a lower socioeconomic position. As to the effect on the transition to parenthood, Corijn and Klijzing (2001) found a negative effect of educational attainment for both for men and women in several Western European countries, but the effect is weaker for men than for women. Yet, in France, the effect for men is positive, but it is U-shaped for women: both low- and highly educated women show higher first-birth rates compared with the medium educated (Winkler-Dworak and Toulemon 2007). Perhaps some of these inconsistencies relate to the fact that the effect of education may change over the life course. Previous studies have suggested that the association between educational attainment and the transition to parenthood may depend on time since graduation (Brien et al. 1999; Martin-Garcia 2009; Martin-Garcia and Baizan 2006; Winkler-Dworak and Toulemon 2007). The economic rationale for this is that differences in earning potential may be evidenced only a few years after graduation, after a professional career has been established, and people get ready for family formation. This would hold most for people with advanced degrees.

In the literature on fertility from a couple’s perspective, scholars have been looking at the relative influence of partners’ characteristics on the transition to parenthood (Begall 2013; Corijn et al. 1996; Gustafsson and Worku 2006; Jalovaara and Miettinen 2013; Martin-Garcia 2009; Thomson 1997; Vignoli et al. 2012). These studies include individuals who were in a coresidential union at the time of data collection, thus implying that those less likely to enter or stay in a union are more likely to stay out of the picture. This is a crucial shortcoming because it disregards the effects of mate selection on fertility. More specifically, we expect that men’s education affects their fathering rates chiefly through its effect on forming cohabiting or marital unions with women. If so, studies from a couple’s perspective might wrongly conclude that the male partner’s education matters less for fertility than the female partner’s; in fact, his education may matter as much or more, but only during a different stage of the family formation process.

Few studies have accounted for the interrelationships between union formation and fertility, applying the simultaneous equations approach for hazards developed by Lillard and colleagues (Brien et al. 1999; Lillard 1993; Lillard and Panis 2003; Lillard and Waite 1993; Lillard et al. 1995). A few studies have analyzed the interrelationship between first union formation and the transition to parenthood for women (Baizán et al. 2003; Brien et al. 1999). Both studies concluded that the two processes share unobserved factors that jointly affect the experience of events. For men, the relationship between union formation and the transition to fatherhood has hardly been studied. Winkler-Dworak and Toulemon (2007) considered the role of men’s selection into unions when analyzing the transition to fatherhood in France. Their results suggest that part of the positive effect of men’s educational attainment on the transition to fatherhood was driven by the higher rate of union formation among highly educated men. In Finland, Jalovaara (2012) and Jalovaara and Miettinen (2013) found that socioeconomic resources for Finnish men and women were important to be selected into unions as well as to become parents. Still, the authors found that the female partner’s education has a stronger impact on the transition to parenthood than the male partner’s education (Jalovaara and Miettinen 2013). The latter finding suggests, in line with our hypothesis, that the man’s education matters more for being selected into a union rather than for becoming a father after being selected into a union.

The Selection-Into-Union Hypothesis

Based on the theoretical arguments and earlier empirical studies thus presented, we expect that the level of educational attainment has a consistently positive effect on men’s transition to fatherhood, but that this effect is largely indirect, namely through its positive effect on the rate of union formation. The underlying assumption is that highly educated men tend to be attractive on the mating market. An economic reason for their attractiveness is their relatively high earning potential. Another attractive feature, at least for some women, is that highly educated men may be more likely to hold egalitarian gender role attitudes and thus may be more prone to share household chores with their partners (Coltrane 2000; Sullivan et al. 2014). Lower-educated men have more difficulty finding a committed partner and therefore, all else being equal, are expected to experience lower fatherhood rates (Kravdal and Rindfuss 2008; Lappegård and Rønsen 2013; Winkler-Dworak and Toulemon 2007).

These expectations hold for men who have completed their studies and who are no longer enrolled in education. The effect of enrollment in education is expected to be negative throughout. So, even if men who pursue a college degree will have their first child later, we are predicting consistently higher fatherhood rates for them after they obtain their higher degree. Our hypothesis implies that the higher fatherhood rates for the college-educated can be explained by the fact that they are able to match with a committed female partner more quickly than their low-educated counterparts. When we model union formation and fatherhood jointly, we expect to find a consistent positive effect of educational attainment on union formation but no consistent effect on the transition to fatherhood.

Thus, the competing hypothesis is that the educational attainment level still has a positive effect on fatherhood rates after accounting for selection into union. At least three potential reasons for such a competing hypothesis could be mentioned. First, higher-educated men may be more likely to enter parenthood even after accounting for union formation because they have been found to be more likely to marry (Jalovaara 2012). Second, some studies on couple-level fertility have suggested that men’s education-related socioeconomic resources may stimulate the couple’s childbearing behavior (Jalovaara and Miettinen 2013; Kreyenfeld and Konietzka 2008; Vignoli et al. 2012). Third, partnered men with advanced education may experience higher fatherhood rates due to lower divorce and separation risks (Jalovaara 2011).

Data and Methods

Sample Selection

To test our hypothesis, we used survey data of the Generations and Gender Surveys (GGS, see http://www.ggp-i.org/) for 10 European countries that provide the information needed: Austria, Belgium, Bulgaria, Estonia, France, Hungary, Lithuania, Norway, Poland, and Romania. We chose to replicate our empirical tests in these 10 countries rather than focusing on just one or a few countries. We included these particular countries because the GGS data needed to test our hypothesis are available for them.¹ The number of countries is insufficient, however, to test the role of country-level factors using multilevel models.

The GGS surveys include men and women aged 18–79 and address topics such as fertility and partnership histories, the transition to adulthood, economic activity, care duties, and attitudes (Vikat et al. 2007). Response rates are above 60 % in all countries except Belgium (41.8 %) and Lithuania (35.6 %) (Fokkema et al. 2016); caution is needed in particular for these two countries. For this study, we selected men born after 1949. Men were censored at age 45 for both union formation and first birth because both first unions and first births very rarely occur at older ages even among men. We used information about the month and year of events. If the month was missing, we randomly imputed it. From an initial sample of 51,224 men (for all countries), we excluded from the analysis men involved in same-sex relationships (n = 163) and those born before 1950 (n = 14,881). We then dropped cases (1) with missing or misreported information on the date of first union (n = 703) as well as date of first birth (n = 28); (2) for whom it was not possible to determine whether the event of interest occurred; and (3) for whom the event occurred before the 15th birthday (n = 125 for first union and n = 29 for first birth). After these selections, our sample totaled 35,295 men. We distinguished between three birth cohorts: 1950–1959, 1960–1969, and 1970–1990.² Table 1 gives descriptive statistics for the samples and variables used.

Measures

We coded the date of first partnership formation using information on the month and year of the first reported coresidential partnership, distinguishing between unmarried cohabitation and marriage and keeping track of any changes in marital status. The GGS surveys collected information only on partnerships that lasted at least three months (Vikat et al. 2007). To focus on the relationship between first union as well as first birth, respondents who experienced more than one coresidential union were censored at the end of the first one. Thus, first births occurring in higher-order unions are not considered. Hence, if no transition to fatherhood is observed, men are censored at (1) age 45, (2) the time of interview, or (3) the break-up of the first union, whichever comes first. Only in Norway and Austria did the proportions of first births in higher-order unions exceed 10 % (12 % and 11 %, respectively). In France, it was almost 9 %; in Belgium and Hungary, 7 %; and in the rest of the countries, approximately 1 %. As a robustness check, for countries where the proportion of first births in higher-order unions exceeded 6 %, we ran a version of our models that was modified to include higher-order unions. The results do not deviate substantially form the ones reported in this article.

We backdated the date of the transition to fatherhood by eight months to avoid anticipation bias (Baizán et al. 2003), based on the date of birth reported by the respondent for his first biological child, if any. Men may underreport their fertility—by a major margin in the United States and the UK (Joyner et al. 2012; Rendall et al. 1999)—with only minor underreporting in Russia (Alich 2009). The underreporting of male fertility is likely to be selective with respect to union status and education: births outside marital or cohabiting unions and births to men with low educational attainment are expected to be disproportionally at risk to remain underreported. Given that low-educated men are likely to be underrepresented in our data because of nonresponse (Fokkema et al. 2016), births to low-educated men are most likely underreported in our data, particularly births to men who are not living with the mother of their first child.

Our analysis looks at two dimensions of education: enrollment and attainment. We constructed both variables as time-varying covariates. To capture the enrollment dimension, we included time since graduation as a categorical variable with three categories: (1) still enrolled in education; (2) up to two years after graduation; and (3) more than two years after graduation. We distinguished between the very recent graduates and those who left school or college more than two years ago because finding a job after graduation takes time (typically one to two years; (Quintini and Manfredi 2009)), followed by earning a living and becoming ready for family formation (cf. Martin-Garcia 2009). The end of enrollment is based on the time of actual graduation as reported by the respondent in most cases (93.6 %). When this information was missing, we imputed the end of enrollment as the standard age at graduation for the relevant attainment level and country. We had to do this for 6.4 % of the cases overall. For men who completed a college level degree, where study duration varies more, the percentage of missing data on time of graduation was only 0.4 %, making the need for imputation for the latter group minimal. Still, as a robustness check, we reran our models after dropping cases where information on the actual date of graduation was missing. The results remained basically the same and would not affect our conclusions.

For educational attainment, we grouped men into three levels (low, medium, and high), collapsing categories from the International Standard Classification of Education (ISCED) 1997. The first group includes those who completed primary plus lower secondary school (at least eight years of schooling: ISCED 0, 1, and 2). The medium category consists of men who attained the upper secondary level and those who reached postsecondary level (ISCED 3 and 4). Finally, highly educated men are those who earned an advanced (bachelor’s/master’s/PhD) degree (ISCED 5 and 6).

Corijn and Klijzing (2001) argued that the effect of education on the transition to parenthood may not be constant but changing over the life course, which would represent a violation of the proportionality assumption. One way to try to account for this time dependency is to interact age with education. However, earlier studies have suggested that the time dependence is a function of time since graduation rather than of age per se (Brien et al. 1999; Martin-Garcia 2009; Martin-Garcia and Baizan 2006), with age at graduation obviously correlating with the level of the degree obtained. We therefore included interaction terms between men’s educational attainment and the three categories of years since graduation in all models. Apart from that, we also included five-year age splines to accommodate any nonlinear relationship between age and the events of interest (see upcoming discussion for more technical details).

We categorized the female partner’s education the same. To catch the effect of a long-term dimension of the social status, we included parents’ educational attainment, coded into four categories (both parents low-educated, only the father at least medium-educated, only the mother at least medium-educated, and both parents at least medium-educated). Because individuals with more siblings are more prone to start a family (Murphy 2013), we included the number of siblings as a time-constant variable.

We included the aforementioned variables in the models for both first birth and first union. The model of first birth includes the time-varying endogenous variable union status, indicating whether the male respondent is living in a coresidential union. For men living with a female partner, we distinguished between those partnered with a low-, medium-, or highly educated woman. We added a category “not available” to accommodate men in union but with missing information on the partner’s education. We also included a time-varying dummy variable indicating whether the union is a formal marriage rather than unmarried cohabitation. Finally, in the model of first union formation, we included a time-varying dummy variable for the conception of the first child (birth date backdated by eight months).

Analytical Strategy

The superscripts F and U refer to the equation for fatherhood and union formation, respectively. The random variables ε and δ represent unobserved heterogeneity terms, which are assumed to have a joint bivariate standard normal distribution:

Given our focus on only first unions and births, we deal with nonrepeatable events. Aassve et al. (2003) showed that the estimates for hazard models for nonrepeatable, correlated events may be sensitive to the variance of the unobserved heterogeneity term. To check the sensitivity in our case, we ran two versions of our models: (1) fixing the variance to 1 (the variance of the standard normal distribution), and (2) estimating the variance empirically. The results tended to be robust to the direction of the effects and their significance, changing slightly with regard to the magnitude of the effects. In France and Estonia, the estimated correlation between unobserved factors changed both in sign and statistical significance, but the estimates for the fixed effects of education—the effects of substantive interest—remained stable in these two countries as well.

The joint modeling approach can account for the correlation between the processes of union formation and entry into fatherhood for various reasons, including causal effects running in both directions. Still, we were concerned about how “shotgun” union formation (i.e., when pregnancy induces union formation) may affect our estimates. As a robustness check, we reran our models in which we backdated all births by 18 months, which implies that only children conceived approximately nine months after union formation were counted as born within the union. The results are in line with the ones reported in this article.

We estimated the models for each country sample separately. We did not apply multilevel modeling because our focus is not to test the effect of country characteristics. Rather, we want to replicate the same hypothesis testing in different contexts. Still, the number of countries would be too small to apply multilevel modeling and test hypotheses about the role of country characteristics. In an additional analysis not reported, we pooled our 10 countries. Those results are an averaged, stylized summary of our main finding, consistent with what we report for countries separately.

Results

We fitted three models for each of the 10 countries. Online Resource 1 reports the estimates for all 30 models. For each country, Model 1 is the model of the transition to fatherhood that does not control for union status. Controls in this model as well as in all subsequent models are age, cohort, time since graduation, own educational attainment of the male respondent, parental educational attainment, and number of siblings. Model 2 adds the control for union status, type of union, and the educational attainment of the female partner, if any. Figure 1 plots the effects of own educational attainment on the transition to fatherhood, estimated from Model 1, along with their 95 % confidence intervals; Fig. 2 shows the same for Model 2. All figures compare the rates for men with high or medium educational attainment with the reference category with low educational attainment (represented by the 1.0 line), two years or more after graduating from school or college.

As shown in Fig. 1, educational attainment has a consistently positive effect on the transition to fatherhood when union status is not controlled for. In all countries except Austria, the difference between low-educated and highly educated men is statistically significant. In the majority of countries, the fatherhood rates are more than 50 % higher for college-educated men compared with men with low educational attainment. The relative rate for medium-educated men is in most countries in between. Figure 2 shows that after we control for union status, the effect of education is strongly reduced in most of the countries. Thus, most of the positive effect of education on the transition to fatherhood appears to be driven by union formation.

Next, we accounted for the selection-into-union effect and the endogeneity of union formation and fatherhood by simultaneously estimating joint models of both processes. See Model 3 in Online Resource 1. Table 2 reports for each country the estimates for the fatherhood equation of Model 3, and Table 3 reports estimates for the union formation equation.

To facilitate interpretation, Fig. 3 plots the effects of male educational attainment on fatherhood (panel a) and union formation (panel b). Generally, educational attainment has no effect on fatherhood rates after we account for the selection-into-union process in most countries. Belgium, Romania, and Bulgaria are the only countries where we still see a direct, statistically significant effect: in these three countries, college-educated men exhibit higher fatherhood rates compared with men with low educational attainment. In Austria, the point estimate for highly educated men is also a bit above the 1.0 reference line of the low-educated men, but the difference is not statistically significant. For medium-educated men, no significant differences are evident in any country. In contrast, educational attainment positively influences union formation in most countries, with statistically significant differences in all countries except Bulgaria and Norway. In France, unexpectedly, medium-educated men appear to exhibit higher union formation rates than highly educated men, but the difference is not statistically significant. All in all, the estimates quite consistently point to a positive educational gradient in union formation rates.

The control variables in all models (Online Resource 1, and Tables 2 and 3 for the joint Model 3) tend to have the expected effects, but discussing these is beyond the focus of this article. The joint model also yields estimates of the correlation between residual, unobserved terms in the fatherhood and union formation equations (see Online Resource 2). Both processes are obviously positively correlated. Indeed, in all countries, the estimated correlation between the unobserved factors in the joint model is positive as long as we exclude any of the two endogenous variables (union status in the equation of first birth or conception in the equation of first union). Interestingly, however, after we include in the system of equations both endogenous variables, the positive correlation between unobserved factors disappears in almost all countries. Only in Belgium does the correlation between unobserved factors remain positive. In Austria, it is not significant (probably due to the small sample size); in Norway, it is positive but not significant; and in France, the significance and sign of the correlation term are sensitive to the values of the unobserved heterogeneity factors.

In all eastern European countries (Bulgaria, Hungary, Poland, Romania, and Lithuania), the correlation between unobserved factors in both processes turns negative; however, the estimated correlation is not robust for Estonia. Such negative correlation implies that after the effect of union formation on fatherhood and the effect of fatherhood on union formation are accounted for, some unobserved factors enhance the experience of one event but delay the other. For instance, perhaps men with unobserved personality traits who are particularly eager to enter a romantic relationship decide to live with a partner relatively early but are not eager to start fathering a child. Testing this interpretation is beyond the focus of this study.

Discussion

Several authors have recently argued that men’s attitudes, intentions, and behaviors are becoming increasingly important factors to understand patterns and trends in family formation (Goldscheider et al. 2014; Huinink and Kohli 2014; Van Bavel 2012). Yet, empirical research so far has not yielded clear and consistent results about how men’s characteristics affect the transition to parenthood. In this article, we argued that the explanation for this may be that men’s characteristics perhaps matter most in the process of union formation, such that men with more attractive features for women are more likely to be selected as their partners. Empirical analyses that look at the couple level, after unions have been formed, will then perhaps fail to see any clear effects of male characteristics, net of the effects of female characteristics.

More specifically, in this study, we tested the selection-into-union hypothesis with a focus on educational attainment. We expected that educational attainment has a consistently positive effect on men’s transition to fatherhood but that this effect is largely indirect, working through its positive effect on the rate of union formation. Our results generally supported the hypothesis, with differences between European countries but with a clear overall pattern: we found a positive educational gradient in men’s union formation but, after accounting for that, not in their transition to fatherhood. This pattern showed up particularly for men who have been out of school for more than two years, presumably the time needed for the majority to have gained an established position in the labor market. Before that—just after leaving school—the results are more mixed, but we found little support for a competing hypothesis that men’s educational attainment would have a direct positive effect on fatherhood. Only in Belgium, Bulgaria, and Romania did we still find a positive effect of high educational attainment on fatherhood compared with low-educational attainment.

Our hypothesis appears to apply particularly strongly in Central and Eastern European countries, including the Baltic countries Lithuania and Estonia. In the latter two countries, we found a clear positive educational gradient in fatherhood rates, but this gradient vanished completely when selection into union was accounted for. The positive educational gradient in union formation emerged even more clearly in the joint modeling framework.

In Western Europe, Belgium is the only country where a positive educational gradient in fatherhood remained in the joint modeling framework, after we accounted for union formation. Here, we can only speculate about the reasons for the Belgian exception, and it remains to be seen whether future studies with other data can replicate our GGS-based finding. If real, a positive educational gradient in fatherhood (after accounting for selection-into-union) could signify that father role expectations are strong and that the highly educated are the most likely to meet these expectations. An alternative explanation could be that the effect of educational attainment represents a direct income effect on the transition to fatherhood. We speculate, however, that this would rather hold for the other two countries where a positive effect of education on fatherhood remains in the joint model—namely, in Romania and Bulgaria, two economically disadvantaged Eastern European countries. In these countries, more highly educated men can perhaps more easily “afford” to father children, and perhaps low-educated fail to report any children they have, if they are aware of them at all.

By jointly modeling union formation and fatherhood, we were able to overcome major limitations of earlier work on education and the transition to fatherhood. Earlier studies with models that included the education of the male partner along with the education of the female partner included only those men who were partnered already. We argue that this approach suffers from a crucial selection bias because men’s education chiefly influences fertility through the selection into unions. That crucial selection effect has been overlooked in studies investigating fertility at the couple level. Alternatively, studies in the transition-to-adulthood tradition have typically failed to control for the female partner’s educational attainment, ignoring strong educational homogamy and, hence, were unable to tell whether it is his or her education that matters.

This study has its limitations. First, we cannot be sure whether the men in our sample had other, perhaps unacknowledged children with other women than the ones identified in our data. Some men may have underreported their children in GGS data, either unintentionally (they simply did not know about those children) or intentionally (Alich 2009). Intentional underreporting will likely be selective with respect to union status and education. Men who are afraid to be socially sanctioned for their extra-union childbearing behavior, or men who are in trouble for paying child alimony, may tend to omit children born from dissolved unions and who do not coreside with them (Joyner et al. 2012; Lindberg et al. 1998). Thus, perhaps births to low-educated men are most likely underreported in our data, particularly births to men who are not living with the mother of their first child. Most likely, however, men will not be engaged in much active fathering of their unacknowledged or unreported biological children. Second, we did not distinguish between marriage and unmarried cohabitation in this study. For example, we did not model the selection process that would lead highly committed men to marry for having children rather than to cohabit. This could affect the correlation terms between unobserved factors: men who signal stronger commitment by marrying would probably have higher fatherhood rates than those who choose to cohabit. As a result, the positive correlation between the transition to parenthood and marriage would be stronger compared with the correlation with unmarried cohabitation. The aim of this study, however, was not to analyze the role of different kinds of living arrangement histories but rather to assess the role played by men’s educational attainment in their transition to parenthood through the selection into union. Although we did not model the selection process into marriage versus cohabitation, we did control for the distinction in the equation for the transition to fatherhood. Third, our study focused only on two dimensions of education: enrollment and attainment. Still, it would be interesting to test whether the selection-into-union hypothesis holds with regard to the effect of educational field of study. Finally, future research could also address the role of mating market composition in terms of educational attainment on the selection effect at the time of union, including both individual- and macro-level indicators.

Even with all mentioned limitations in mind, we argue that the selection-into-union effect should be taken into account, particularly when comparing the role of education in the transition to parenthood between men and women. Currently, the consensus in the field seems to be that women’s education matters more for fertility than men’s. However, in this article, we show that earlier research may have strongly underestimated the role of men’s education because the process of their selection into unions has been ignored.

With the reversal of the gender gap in education, the selection of men into unions based on their education may even become a more important factor, given that the growing numbers of college-educated women typically are looking for partners with the same educational status (Van Bavel 2012). More generally, we speculate that role of men’s characteristics, intentions, and behaviors may become more and more important for future fertility trends and patterns because college-educated women are increasingly competing for men with similar education. The relative scarcity of the latter on the marriage market may enhance their power to have their say in decision-marking about fertility. But researchers who fail to account for the selection-into-union process risk missing that point.

It would be interesting to compare the effect of education on fertility through selection-into-union between men and women. Perhaps selection on education is playing an increasingly important role for women as their contributions to the household budgets become more significant and as their own earning potential plays an increasingly important role in mate selection by men. Several studies have reported that the positive effect of education and its earning potential on marriage rates has strongly increased over time (Oppenheimer 1994, 1997; Sweeney 2002). In a recent European study, low-educated mothers were found to remain single more often than in earlier cohorts and compared with college-educated women (De Hauw et al. 2015). Selection based on education may therefore increasingly play a role for women’s transition to parenthood as well. These findings and trends highlight the relevance of investigating the selection-into-union hypothesis not only for men, as in this study, but also for women.

Acknowledgments

The research leading to these results has received funding from the European Research Council under the European Union’s Seventh Framework Program (FP/2007-2013)/ERC Grant Agreement No. 312290 for the GENDERBALL project.

Notes

References