how far can the shift to later childbearing in developed countries be accounted for
by the growth in educational participation? A move to later childbearing has been
a conspicuous feature of fertility trends in developed countries for many decades,
and over that same period educational participation rates have risen substantially
(OECD 2014, 2016a). The rising age at birth is often described as fertility postponement
and is primarily due to a progressively later start to childbearing. Fewer women have
been starting a family in their teens and early 20s and more have been delaying the
start of parenthood to their late 20s and 30s (d'Addio and Mira d'Ercole 2005; OECD
2016b). The consequence is a decline in the first birth rates of childless women at
younger ages, followed, in most cases, by a rise in parity‐specific rates at older
ages, resulting in a general shift up the age scale in the timetable of parenthood.
In the West, the mean age at first birth began rising in the 1970s; in Eastern Europe,
the trend began in the 1990s, following political and societal transformation (see
Figures 1a and 1b). Later childbearing is emerging more recently as a feature of population
trends in Southeast Asia and Latin America (Rosero‐Bixby et al. 2009; Frejka et al.
2010).
Figure 1a
Period mean age at first birth, Western Europe and other developed countries, 1960–2014
Figure 1b
Period mean age at first birth, Eastern Europe, 1960–2014
NOTE: Figures show the standardized period mean age at first birth, termed MAB1 in
the Human Fertility Database (Jasilioniene et al. 2015).
SOURCE: Human Fertility Database. www.humanfertility.org (data downloaded 15 January
2016).
The postponement of childbearing in developed countries has been sizable and sustained.
This has demographic importance for a number of reasons. Change in the mean age at
first birth makes time trends in fertility harder to interpret. The indicator most
widely used to measure such trends, the period total fertility rate (TFR), is influenced
by shifts in fertility tempo. Tempo change increases the difficulty of assessing the
extent to which trends in the TFR reflect change in the level of fertility or only
in its timing, or in some combination of the two. Considerable technical ingenuity
and debate have been devoted to this problem in the last two decades (see especially
Bongaarts and Feeney 1998 and associated commentary; Schoen 2004; several chapters
in Barbi et al. 2008; Ní Bhrolcháin 2011). The postponement of fertility also raises
practical issues. Forecasters, policymakers, and planners need an assessment of how
long the postponement phenomenon is likely to continue and its implications for long‐range
levels of fertility, considered from either a period or a cohort perspective. Social
commentators and medical professionals have raised concerns about the implications
of a delayed start to childbearing for women's ability to have children at later ages
(Menken 1985; Billari et al. 2007; te Velde et al. 2007; Schmidt et al. 2012). Delayed
childbearing also has implications for intergenerational support: for example, children
born later are less likely to have surviving parents at any given age, and surviving
parents are likely to be in poorer health (Murphy et al. 2006).
A wide array of potential causes of the shift to later childbearing have been proposed.
For example, a recent review (Mills et al. 2011: 848) summarized the principal factors
influencing aggregate postponement as: “effective contraception, increases in women's
education and labour market participation, value changes, gender equity, partnership
changes, housing conditions, economic uncertainty and the absence of supportive family
policies” (see also Blossfeld and Huinink 1991; Lesthaeghe 2001; Billari et al. 2006).
While there has been no shortage of candidate causal factors, with educational expansion
prominent among them, there have been fewer attempts to quantify the contribution
of specific determinants to aggregate change in fertility timing (but see Rindfuss
et al. 1996; Bergouignan 2006; Neels 2009; Neels and De Wachter 2010; Ní Bhrolcháin
and Beaujouan 2012).
A strong link at the individual level between education and family formation has been
documented in several decades of demographic research. In particular, copious micro‐level
evidence has been available for several decades that better‐educated women in developed
countries start childbearing at later ages than the less well educated (Martin 2000;
Gustafsson et al. 2002; Rendall et al. 2005; Kravdal and Rindfuss 2008). Although
the effect of childbearing on finishing education is likely to contribute to the negative
association between education and timing of parenthood (Cohen et al. 2011; Gerster
et al. 2014), the frequency with which the end of education coincides with a first
birth is limited, indicating that other mechanisms are at play. Education has mostly
been interpreted as influencing family formation either through economic processes—the
career incentives and opportunity costs of childbearing that result from higher attainment—or
through cultural ones—via the acquisition of knowledge and attitudinal and value change
that accompany educational participation. But the time cost of educational participation
and its impact on the age at leaving full‐time education, as distinct from attainment
per se, is a further mechanism by which education may have an effect on the timing
of life‐course transitions (Hogan 1978; Marini 1985; Blossfeld and Huinink 1991; Oppenheimer
1994).
Up to the 1980s, most of the empirical evidence related to the link between education
and family formation was based on measures of educational attainment, such as highest
level of qualifications or years of education. This may have been partly because attainment
was the principal information collected on education in demographic surveys and that
questions were not usually asked on the age at leaving school or college/university.
In addition, conventional regression methods are not well suited to examining the
statistical effects of attributes that vary within an individual life course, such
as educational enrollment, employment status, and the like. With the advent in the
1980s of methods of event history analysis that allowed the investigation of time‐varying
covariates, it became possible to examine the effect of time spent in education per
se as distinct from ultimate attainment or highest level of qualifications.1 Numerous
studies have investigated the rate of transition to adult statuses—cohabitation, marriage,
first birth—using this methodology. Probably the most consistent finding is that first
birth rates are substantially lower when women are enrolled in education (Blossfeld
and Huinink 1991; Blossfeld and Jaenichen 1992; Kravdal 1994; Liefbroer and Corijn
1999; Santow and Bracher 2001; Winkler‐Dworak and Toulemon 2007). By contrast, the
statistical effect of educational attainment on rates of first birth is much weaker
or often absent when enrollment status is taken into account (Blossfeld and Huinink
1991; Santow and Bracher 2001; Lappegård and Rønsen 2005). The question arises therefore
as to the extent to which the link between educational attainment and adult transitions
is attributable to i) variation in time spent in education and the corresponding effect
on the age at which people leave full‐time education and ii) to other aspects of educational
attainment, such as its impact on earning power, attitudes and values, and participation
in the labor force (Marini 1985). We examine the first of these issues in this article,
evaluating the extent to which the changing age at first birth is linked to the change
over time in educational enrollment and to the resulting upward shift in the age at
leaving education.
One cannot infer from the consistent evidence of an individual‐level link between
education and birth timing that the growth in educational participation explains the
trend over time to later childbearing. The micro‐level link may not be a causal one,
and so even if both education and birth timing change over time in a corresponding
direction, this may not be due to a causal process. And even if education is a cause
of fertility delay at the micro level, the causal effect may not be strong enough
to account for any observed aggregate change in birth timing. Although rising education
levels have often been suggested as a potential cause of fertility postponement, only
a handful of studies have investigated the degree to which changing educational composition
accounts for delayed marriage or first birth in developed countries. Using standardization,
Rindfuss et al. (1996) found that the changing distribution by educational attainment
did not explain the rise in age at first birth in the US. Their results appear anomalous,
given the data and differentials reported.2 By contrast, two standardization analyses
of Belgian data have shown that rising attainment accounts for 37–54 percent of the
cumulated deficit in the proportion of women having a first child by age 25 in the
cohorts born between 1951 and 1975 relative to the 1946–1950 birth cohorts. Similarly,
adopting a period perspective, two‐thirds of the rise in mean age at first birth between
1970 and 2000 is accounted for by rising levels of educational attainment over the
period considered (Neels 2009; Neels and De Wachter 2010). Only two studies have analyzed
the effects of compositional change in educational enrollment or the age at completing
education on delaying the entry to parenthood, both using standardization. Bergouignan
(2006) reported that about half of the increase across the female French cohorts of
1960 and later in the proportions childless at ages 25 and 30 was attributable to
the rise in the age at leaving education. Ní Bhrolcháin and Beaujouan (2012), adopting
a period life table approach to standardize for changing structure by age and by educational
enrollment/duration, estimated that around three‐fifths of the rise in the 1980s and
1990s in the period mean age at birth in Britain, and four‐fifths in France, was attributable
to later ages at completing education and, thus, to rising educational enrollment.
Over the same period, the increase in the time to first birth after completing education
was found to be greater among better‐educated women than among the less well educated.
We extend the investigation of Ní Bhrolcháin and Beaujouan in several ways. The standardization
for enrollment and duration since leaving education is more comprehensive than in
the earlier study. We improve on the previous estimates by adopting a modeling approach
to smooth the estimated rates. We examine the contribution of these structural factors
to change in the mean age and to shifts in age‐specific first birth rates. In addition,
we separate the overall structural contribution to period change in first birth timing
into two components—one due to rising enrollment at each age, the second to the changing
composition at each age by duration since leaving education. We add a further data
source for the UK, allowing us to make estimates back to the 1970s. Finally, we add
an additional country to the cross‐national study, using large‐scale data from the
Belgian census of 2001.
Data
For the United Kingdom we used two survey sources: (a) a pooled series of General
Household Survey (GHS) rounds from 2000 to 2009, on which the Ní Bhrolcháin and Beaujouan
(2012) analysis was based; this is a subset of a larger harmonized time‐series data
file of GHS surveys from 1979 to 2009 compiled by the Centre for Population Change
(Beaujouan et al. 2014a); (b) to this dataset we added the first (2009) round of the
UK Household Longitudinal Study (UKHLS), also known as Understanding Society, a prospective
longitudinal study linked with and incorporating the British Household Panel Survey
(Knies 2015). Both surveys collected fertility histories, the GHS for women aged 16–59
and UKHLS for women aged 16+. Both also fielded questions on the age at which respondents
completed their education, discussed further below. Data from the two sources were
pooled to form a single database and validated against period fertility rates derived
from the national vital statistical system and against national statistics on educational
participation. We confine analysis of GHS data to the later rounds, 2000–2009, because
the information on the age at finishing education was found to be defective before
the 2000 GHS round (Beaujouan et al. 2014a, Appendix G).3
For France, we used the same large‐scale survey data as in the earlier study, the
Family History Survey (FHS/EHF) linked with the French census of 1999 (Cassan et al.
2000). This self‐completion survey collected details on the fertility histories of
women aged 18 and above and on the age at completing education, discussed further
below. The FHS/EHF, the main source of data on parity‐specific fertility in France,
has been validated and weighted by INSEE and INED (Mazuy and Toulemon 2001).
Finally, we use the fertility histories of women aged 14+ together with information
on age at achieving highest qualification collected in the 2001 Belgian census (Deboosere
and Willaert 2004), on which the analyses by Neels (2009) and Neels and De Wachter
(2010) were based. Validation against vital registration has shown that period and
cohort indicators estimated retrospectively from the 2001 census agree well with national
statistics on period fertility trends between 1960 and 2000 and with independent estimates
of cohort fertility patterns for women born after 1930 (Neels and Gadeyne 2010).4
Month and year of the woman's first birth were available in all four data sources
used. Both month and year of first birth were collected in the UK and French surveys,
but since only the year of first birth was available in the Belgian census we assigned
a birth month of June to all first births in this source.
Data are available up to 2000 for the UK and Belgium. The FHS took place in 1999 but
for technical reasons the French data are available only up to 1998. We chose 1970
as a starting point both because it precedes by a few years the start of the lengthy
rise in age at first birth in all three countries and because UK data before about
1970 become increasingly selected since they are mainly based on retrospective information
from respondents alive in 2000–2009 and, in the case of the GHS, aged under 60. We
therefore confine analysis to the period 1970–1998/2000.
Age at completing education
The age at which individuals complete their education is not readily defined. People
may leave school, engage in some other activity such as employment or family formation,
return later to full‐time education, and exit again. Thus, they may leave full‐time
education several times during their lives. We therefore defined our key variable
as the age at which a person first left continuous, full‐time education and attempted
to construct this. The indicator remains fixed throughout a person's lifetime, unlike
educational attainment or total years spent in education. That in turn makes it much
less prone to problems of endogeneity and reverse causation than highest educational
qualification or years of education (Kravdal 2004; Hoem and Kreyenfeld 2006). Accurate
measurement of our key indicator would require a near‐complete dated history of the
start and end of spells of full‐time education. Such information was not available
to us and is rarely collected. Details of how the indicator was constructed in each
dataset are given in the online Appendix.5 For economy we refer to this variable throughout
as age at leaving education or age at completing education.
Methods
Our fertility indicators are annual age‐specific first birth rates among childless
women aged 15–39 who were in education while below age 26, or reported leaving full‐time
education between ages 15 and 26, and the period life table mean ages at first birth
derived from these schedules. To measure the propensity to leave school by age in
a calendar year, we use a period life table mean age at leaving education, calculated
analogously to the period life table mean age at first birth. Both of these measures
are independent of the age structure in a given year and are therefore free of the
distortions to which crude period mean ages are subject when age structure varies
over time. We refer to these measures throughout as period mean age at first birth
and period mean age at leaving education.
We decompose age‐specific first birth rates in a given year into the weighted sum
of i) the age‐specific first birth rates of women enrolled in education, and ii) the
age‐specific first birth rates of women who have left education. Moreover, since the
fertility behavior of the latter women is also determined by the duration since leaving
education,6 we additionally disaggregate the age‐specific rates of women who are out
of education into a weighted sum of age‐duration‐specific first birth rates. Details
of the decomposition are provided in the online Appendix.
We use indirect standardization to assess the extent to which changing educational
participation accounts both for the rise in the mean age at first birth and for the
change in the first birth rates of childless women at each age. In each case, we estimate
two distinct but related structural effects:
(i) We evaluate the impact of changing enrollment on change in the mean age at first
birth and on change in age‐specific first birth rates, by standardizing for age‐specific
enrollment.
(ii) Change in educational enrollment at each age results in a change in the distribution
of ages at leaving education, and this in turn results in change in the composition
of each age group by duration since leaving education. We estimate the impact of changing
duration by standardizing jointly for enrollment and duration since leaving education,7
and taking the difference with (i).
Our preferred standard schedule of rates is the average of the annual age‐specific
rates of women in education and the average of the annual age‐duration‐specific rates
of women who left education across the 29/31 years from 1970–1998/2000. We consider
this the most appropriate standard for the present purpose as it minimizes the deviations
from the actual values and, given significant changes in rates across the period analyzed,
alternative standards could differ substantially from some observed values. However,
because results can vary with the standard employed, we also use two other standards—rates
in 1970, the start of the period, and rates at the end of the period, 1998/2000—to
assess the sensitivity of the estimated effects. Details of the standardization are
given in the online Appendix.
The standard schedules used in the indirect standardization were smoothed using generalized
additive models (details of the generalized additive model (GAM) fitting process are
given in the online Appendix; see also Wood 2006). The smoothed rates improve on the
unsmoothed estimates when samples are restricted in size. Further details are given
in the online Appendix.
Results
Between 1970 and 1998/2000 the mean age at first birth increased substantially in
Britain, France, and Belgium. The period mean age at first birth in 1970 was similar
in the three countries: 23.9 years in France and Belgium and 24.7 years in Britain.
By the turn of the century the figure had risen to 27.2 (UK), 27.5 (France), and 27.1
(Belgium). Thus, the period mean age at first birth rose by 2.5, 3.6, and 3.2 years
in the final three decades of the twentieth century. These substantial increases reflect
the postponement of family formation that was such a striking feature of childbearing
trends in the late twentieth century and that continues up to the present.
In parallel with these shifts was a substantial expansion in educational participation,
with young women in all three countries completing their education at progressively
later ages between 1970 and the turn of the century. The period mean age at first
leaving continuous full‐time education was 17.6 in the UK in 1970 and had risen to
19.6 by 2000. In France and Belgium, the corresponding increases were somewhat larger,
with the period mean age at completion rising, respectively, from an average age of
18.8 in 1970 to 21.5 in 1998 and from 18.9 in 1970 to 21.4 in 2000. During the final
decades of the twentieth century, then, young people were staying on longer in full‐time
education and, as a result, leaving education at progressively later ages. The increases
in the age at leaving education—with young women being between 2.0 and 2.7 years older
on leaving education at the turn of the century compared with three decades earlier—are
sizable enough in principle to lead to substantial shifts in the timetable of childbearing.
Given the well‐established restriction in birth rates during periods of educational
enrollment, and the individual‐level link between educational attainment and later
childbearing reviewed earlier, the question arises whether the growth over time in
educational participation can account for the aggregate trend to later childbearing
and, if so, to what extent.
The beginnings of an answer to this question can be seen graphically, in the contrast
between the left and right columns of Figure 2. The left‐hand plots show smoothed
age‐specific first birth rates of childless women from the 1970s to the late 1990s
for the three countries (original unsmoothed values are also plotted). In each case,
we see a rightward shift in the curve of first birth rates by age. That is, across
these decades the schedule of entry to parenthood moved up the age scale. An alternative
perspective is given by the right‐hand graphs, showing the smoothed first birth rates
of childless women by time since leaving education, again from 1970 onward (we do
not consider births while in education for reasons set out in endnote 6). First birth
rates are closely related to time since leaving education, gradually rising to a peak
and falling again. But the plots by time since leaving education display much less
rightward movement over the three decades than do plots by age. That is, there is
much less change over time in the timetable of first births relative to the end of
education than there is relative to age. The contrast between these two sets of plots
suggests that the upward trend in age at leaving education has played a part in the
aggregate shift to later ages at first birth.
Figure 2
Schedules of first birth rates for women aged 15–39 who have left education, by age
(left‐hand side) and by duration since leaving education (right‐hand side), United
Kingdom 1970–2000, France 1970–1998, and Belgium 1970–2000
NOTE: Schedules are smoothed using generalized additive models (see online Appendix).
SOURCE: General Household Survey and Understanding Society (UK); Family History Survey
and 1999 census (France); 2001 census (Belgium).
Both sets of graphs also show a progressive lowering of the schedules of rates, indicating
a decline in the overall level of period fertility. A conspicuous feature of the trends
during the 30‐year period covered is not only a delay in motherhood, but also a decline
overall in the propensity to start a family. However, because our focus is on explaining
the change in first birth timing, we do not attempt to account for the changing level
of period fertility.
We use indirect standardization to assess the extent to which compositional change
by education can account for the upward shift in ages at first birth. We standardize
for education in two stages. First, we standardize for educational enrollment to evaluate
the impact of the additional time spent in full‐time education. Second, we standardize
for both enrollment and duration since leaving full‐time education. This second stage
is essential to evaluating the total impact of change in educational participation.
The reason is twofold: (1) rising enrollment results in a shift to later ages at leaving
education, and this in turn results in a change in the composition of each age group
by duration since completing education (or, equivalently, by age at leaving education);
and (2) first birth rates are closely linked to time since leaving education, as we
saw in Figure 2.
Results are summarized in Table 1. Between 1970 and 1998/2000 the period mean ages
at first birth rose by 2.5, 3.6, and 3.2 years, respectively, in the UK, France, and
Belgium. We focus mainly on the decomposition based on our preferred standard, the
average 1970–1998/2000 rates in the second panel of the table (see Methods section).
If age‐specific and age‐duration‐specific first birth rates of childless women had
been fixed at their average values throughout the period, but full‐time educational
enrollment rose as observed over the period, the mean ages at first birth would have
risen by 0.7, 1.4, and 1.0 years in the UK, France, and Belgium. That is, the increase
in time spent in education accounts for 26.9, 39.0, and 30.9 percent, respectively,
of the change in the mean age at first birth over the period in the three countries.
Beyond time spent enrolled in education, the changing composition by duration since
leaving education accounts for a further and substantial 1.1, 1.0, and 0.8 years of
the rise in mean age, that is 46.8, 27.3, and 25.9 percent of the increase in the
UK, France, and Belgium. The total structural effect, encompassing changing composition
in both respects, accounts for an increase of 1.8, 2.4, and 1.8 years in the mean
ages at birth. That is, structural factors are responsible overall for 73.7, 66.2,
and 56.8 percent of the aggregate postponement of first births from 1970 to 1998/2000.
Rising educational participation has clearly had a substantial impact on the timetable
of childbearing in the final decades of the twentieth century in these three European
countries.
Table 1
Contribution of rising enrollment and changing distribution of duration since leaving
education to change in period mean age at first birth, estimated by indirect standardization:
United Kingdom 1970–2000, France 1970–1998, and Belgium 1970–2000
United Kingdom
France
Belgium
Period mean age at first birth
1970
24.7
23.9
23.9
1998/2000
27.2
27.5
27.1
Change 1970–1998/2000
2.5
3.6
3.2
Structural effects based on average 1970–1998/2000 standard
Change in mean age due to
Enrollment
0.7
1.4
1.0
Duration since leaving
1.1
1.0
0.8
Enrollment + duration
1.8
2.4
1.8
Percent of overall change explained by
Enrollment
26.9
39.0
30.9
Duration since leaving
46.8
27.3
25.9
Enrollment + duration
73.7
66.2
56.8
Structural effects based on 1970 standard
Change in mean age due to
Enrollment
0.7
1.2
0.9
Duration since leaving
0.7
0.5
0.5
Enrollment + duration
1.4
1.7
1.4
Percent of overall change explained by
Enrollment
26.8
33.1
28.8
Duration since leaving
28.8
13.6
13.0
Enrollment + duration
55.6
46.7
41.8
Structural effects based on 1998/2000 standard
Change in mean age due to
Enrollment
0.9
1.2
1.0
Duration since leaving
1.3
1.1
1.2
Enrollment + duration
2.2
2.3
2.2
Percent of overall change explained by
Enrollment
36.1
33.1
31.7
Duration since leaving
53.2
31.2
34.9
Enrollment + duration
89.3
64.3
66.6
NOTE: Period mean ages at first birth are life‐table adjusted mean ages; the indicator
is thus standardized for age, to abstract from the change over time in the age distribution
of the population at risk of first birth.
SOURCE: General Household Survey and Understanding Society (UK); Family History Survey
and 1999 census (France); 2001 census (Belgium).
John Wiley & Sons, Ltd.
How robust are the estimates? The third and fourth panels of Table 1 show the results
of using two alternative standards—1970 rates and 1998/2000 rates—in the indirect
standardization. The 1970 standard gives a somewhat smaller role to structural change
than does the average standard, while the 1998/2000 standard gives it a slightly larger
role (except in France). The range of the estimated structural effect across all three
standards is 47–66 percent in France and 42–67 percent in Belgium, in both cases indicating
a very sizable effect: at a minimum, structural factors account for close to half
of the timing shift in the two countries. The range of the estimates in the UK, 56–89
percent, is somewhat wider. This may be due to the much more pronounced change in
the shape of the first birth schedule in the UK seen in Figure 2 and may indicate
that standardization is not as effective a method of arriving at a decomposition for
the UK as it is for France and Belgium. However, for reasons we set out earlier, the
average of rates across the period is our preferred standard.
Components of changing age‐specific first birth rates
The mean age at first birth, though indispensable as an indicator of fertility timing,
conveys limited information. A fuller picture is given by examining the entire schedule
of first birth rates by age. Figure 3 decomposes change between 1970 and 1998/2000
in first birth rates at each year of age into structural and rates components, using
the same indirect standardization approach as for the analysis of mean age.
Figure 3
Change between 1970 and 1998/2000 in age‐specific first birth rates: Observed change,
overall structural effect, and rates effect, United Kingdom, France, and Belgium
NOTE: A detailed discussion of the calculation of the observed change, the overall
structural effect, and the rates effect is provided in the online Appendix.
SOURCE: General Household Survey and Understanding Society (UK); Family History Survey
and 1999 census (France); 2001 census (Belgium).
The shift up the age range of the first birth schedules seen in Figure 2 results from
a decline in fertility rates at younger ages and an increase at older ages. This differential
pattern by age is evident for all three countries in Figure 3, which shows substantial
declines in rates at younger ages and modest increases at older ages. The decomposition
reveals that the changing structure by enrollment and by duration since leaving education
contributed to both of these shifts in all three countries. The structural effect
is decidedly negative at younger ages and turns positive at older ages. The three
countries present a similar picture, although the precise ages affected by compositional
change, and the size of the structural effects at each age, differ somewhat between
countries.
Figure 4 shows the enrollment and duration effects separately by age. As would be
expected, the enrollment effect is prominent and negative at younger ages—teens and
early 20s—the ages at which educational participation increased the most.8 The restraining
effect of the duration component is at its strongest around the mid‐20s. The duration
effect then turns positive from the late 20s (France, Belgium) or early 30s (UK) through
the late 30s. Because they left education at later ages, women in their late 20s and
30s in 1998/2000 were out of education for a shorter time than their counterparts
in 1970. First birth rates are, as we saw in the right‐hand graphs in Figure 2, closely
tied to duration since leaving education. Women at these later ages in 1998/2000 were
thus subject to the higher birth rates associated with shorter average durations since
leaving education compared with women of the same age in 1970. Again, the precise
ages affected differ somewhat between countries, but the broad picture is similar
in all three.
Figure 4
Decomposition of the age‐specific overall structural component into duration and enrollment
effects: United Kingdom, France, and Belgium, 1970–1998/2000
NOTE: A detailed discussion of the calculation of the observed change, the overall
structural effect, and the rates effect is provided in the online Appendix.
SOURCE: General Household Survey and Understanding Society (UK); Family History Survey
and 1999 census (France); 2001 census (Belgium).
Compositional factors do not account for the whole of the movements in age‐specific
rates in any of the countries studied. Changing status‐specific rates, reflecting
changing behavior influenced by factors other than educational participation and time
since leaving, also play a sizable role (Figure 3). In France and Belgium, the rates
effect is negative at younger ages. Thus, as well as changing structure (composition),
a declining propensity to start a family among women at an equivalent stage of their
(post) educational trajectory has contributed to the decline in first birth rates
at younger ages in these two countries. In the UK, by contrast, the rates effect is
estimated to be positive at ages under 21. At older ages the effect due to the rates
effect is, like the structure effect, positive in all three countries. Hence the upward
shift in first birth rates at older ages is due not only to structural factors but
also to a rise in the propensity to start a family at these ages. This may, in turn,
be partly attributable to lesser selectivity in the later period. When, as in 1998/2000,
childbearing starts later, those still childless in their late 20s and 30s are a larger
fraction of the cohort than when, as in 1970, childbearing starts earlier. And, because
older childless women are a less select group in 1998/2000 than in 1970, proportionately
fewer of them would be expected to have problems of fecundity than in the earlier
period.
The age profiles of the structural and rates effects are distinctive to each country,
and result from the extent of shifts in enrollment by age, the level of rates at each
age in 1970, and change in age‐ and status‐specific rates. While further analysis
of these differences is beyond the scope of this article, the rates component in all
three countries tends to follow the structural component with a lag of a few years
of age (Figure 3). At younger ages, the rates component tends to reach its largest
negative value about two years of age after the structural component does so; at older
ages, the rates effect tends to reach its maximum positive impact a few years after
the structural effect peaks. This may be purely a statistical effect, due to the fact
that birth rates are rising during the teenage years and falling during the 30s. First
birth rates are therefore lower in the early than in the later teens, and the absolute
size of the rates effect is more constrained at the youngest ages than a few years
later; similarly, the lower rates at older ages toward the end of the fertility schedule
leaves more scope for a larger rates effect. On the other hand, there may be a substantive
basis for the finding, since first birth rates also vary with duration spent in states
that young adults typically experience upon leaving education, such as entry into
the labor market or into a co‐residential union. Hence, the lag may reflect economic
and cultural consequences of rising educational attainment, and particularly variation
in associated labor market, economic, and cultural contexts.
Discussion
We estimate that the rise in educational enrollment accounts for between 57 percent
and 74 percent of the increase in the mean age at first birth in the UK, France, and
Belgium from 1970 to 1998/2000. Although sensitive to the standard employed, the results
indicate that compositional change due to educational expansion has had a sizable
effect on the postponement of first births. The structural effect we estimate for
the UK, at 74 percent, is substantially higher than the 57 percent estimated by Ní
Bhrolcháin and Beaujouan (2012), and our current estimate for France (66 percent)
is somewhat below the earlier 79 percent estimate (the earlier study covered the period
1980–84 to 1995–99). For Belgium our figure of 57 percent falls between the structural
components of 37–54 percent and 66 percent from the analyses of Neels and De Wachter
(2010) and Neels (2009), respectively, employing somewhat different indicators and
approaches. Overall, however, while precise estimates, methods, and, in some respects,
data, differ, our findings corroborate those of these earlier studies—that increasing
educational participation is the primary factor underlying delayed childbearing from
1970 to 1998/2000, accounting for at least half of the fertility postponement over
those decades. The observed change in educational participation, combined with fixed
first birth rates at equivalent stages of the educational trajectory, explains over
half of the postponement observed.
We extended our earlier investigations by separating the structural effect into two
subcomponents—time spent in education and duration since leaving education—and quantifying
their contribution, and also by applying the decomposition to changes in rates at
single years of age. Educational expansion has had an effect on first birth rates
not only via the additional time spent in full‐time education, but also by delaying
the transition to adult statuses and in the process altering the distribution of the
duration since leaving education at each age. Finally, we have shown that compositional
change due to rising education has contributed not only to fertility postponement
but also to fertility recuperation at older ages.
Causation
To what extent do the structural effects of rising educational participation reflect
a causal influence on the aggregate postponement of fertility? And, if a causal process
is involved, through what mechanisms does it operate? Several kinds of evidence are
available on the causal link between education and fertility timing, including both
experimental randomized controlled trials (Mason‐Jones et al. 2016) and observation‐based
studies. Using longitudinal data, a number of studies have shown that the link between
education and later childbearing remains when controlling for family environment and
parental characteristics (Marini 1978, 1985; Rindfuss et al. 1988; Thornton et al.
1995). More explicit evidence of causation is given by several demographic and econometric
analyses of natural experiments, designed to remove the potential endogeneity of education
in relation to fertility. These studies cover a range of countries and exploit a variety
of conditions—administrative rules regarding school entry, changes in the length of
compulsory schooling, change in the duration of vocational education, and the timing
of college entry. These studies are near unanimous in reporting that additional schooling
or a later age at completing education has a causal effect either in delaying the
first birth or, equivalently, in lowering the probability of a teenage birth (Skirbekk
et al. 2004; Black et al. 2008; Monstad et al. 2008; Geruso et al. 2011; Silles 2011;
Cygan‐Rehm and Maeder 2013; Gronqvist and Hall 2013; Humlum et al. 2014; Wilson 2017).9
The estimated causal effects reported in these studies are sizable. Skirbekk et al.
(2004), using a natural experiment framework for Sweden (discussed further below),
found that an 11‐month earlier age at leaving education gave rise to an average 4.9‐month
earlier age at first birth. A range of econometric studies found that an additional
year of schooling reduced the likelihood of a teenage birth by between 3.7 and 8.8
percentage points in the United States and Norway (Black et al. 2008; Monstad et al.
2008), by 5.7 percentage points in Germany (Cygan‐Rehm and Maeder 2013), and by between
4 and 6.1 percentage points in the UK (Silles 2011). A recent Cochrane review of randomized
controlled trials of school‐based interventions related to sex education found that
intervention to retain young people in secondary school reduced adolescent pregnancy,
providing strong evidence for a causal relationship (Mason‐Jones et al. 2016).
Another approach to investigating causality in the relationship between education
and age at first birth is through family‐based designs, especially twin studies. These
examine the extent to which unobserved family‐level factors, environmental and/or
genetic, influence both education and fertility, without a direct link between the
two, and whether there is a direct link from education to age at first birth, independent
of such unmeasured factors. A number of twin and family studies have examined the
relationship between education and fertility. Several of these report that the link
between education and fertility timing is either wholly or largely attributable to
family background or genetic influences (Neiss et al. 2002; Rodgers et al. 2008; Tropf
and Mandemakers 2017). However, these three studies have weaknesses in both methodology
and data. Neiss et al. (2002) use the National Longitudinal Study of Youth to analyze
sibling rather than twin pairs; in addition, the study infers sibling relationships
retrospectively from reports of household membership, includes pairs of siblings thus
identified regardless of sex, and finds no difference between the age at first birth
of brothers and sisters. The quality of data in this study appears insufficient to
bear the weight of the conclusions drawn. The type of twin‐based analysis used by
Rodgers et al. (2008) cannot identify a direct effect from education to fertility
and so cannot illuminate the issue of causation between education and fertility timing.
Finally, Tropf and Mandemakers (2017) do not correct for measurement error, which
can account for a sizable fraction of the within‐pair variance in twin studies; their
estimate of the effect of education on age at first birth is therefore likely to be
biased toward zero (Kohler et al. 2011: 102–104; Amin and Behrman 2014: Section 4.3).10
In contrast, and consistent with the evidence from natural experiments, two recent
twin studies—Nisen et al. (2013) and Amin and Behrman (2014)—find a direct causal
link from education to later childbearing among female co‐twins, net of family and
genetic effects. Using a sample of 628 identical twin pairs from the Minnesota Twin
Registry born 1936–55, Amin and Behrman (2014) found a substantial causal effect,
with one year of additional education leading to fertility postponement of around
one year, a fixed‐effects estimate that was no different from the OLS model.
Some evidence of reverse causation or a feedback effect from fertility to education
has been reported. Cohen et al. (2011) and Gerster et al. (2014) focus on completed
fertility rather than on fertility timing. Furthermore, their results relate to Nordic
populations in which educational careers can be much more extended than in our countries.
The age distribution of full‐time students in tertiary education extends to much later
ages in Nordic countries than in the UK, France, and Belgium; the 85th percentile
age of full‐time students in tertiary education in Nordic countries was between 29.0
and 33.8 in 2006 compared with a range of 23.2–26.4 in the UK, France, and Belgium;
medians are 22.9–24.9 compared with 19.9–20.7 (Eurydice/Eurostat 2009: Figure C17).
In all, a range of investigations, while not unanimous, provide strong evidence of
a sizable causal effect of education on fertility timing. Further, the effect sizes
estimated in these studies are large and consistent in magnitude with the substantial
aggregate effect we report in the present study.
Mechanisms
We suggest that education affects first birth rates at each age in three ways: (1)
the direct impact of factors that inhibit fertility during full‐time educational enrollment;
(2) the displacement to later ages of the timetable of transition to adulthood; and
(3) effects of educational participation on factors such as earning power, labor force
participation of women in particular, attitudes, values and preferences, social learning,
and on knowledge of and access to contraception, beyond the role such factors may
have in the first two pathways. Our estimates are confined to the first two of these
mechanisms, with the first accounting for between 27 and 39 percent of the postponement
of first birth between 1970 and 1998/2000 in the three countries considered, and the
second accounting for 26 to 47 percent of the rise in the mean age at first birth.
But educational expansion may clearly also have effects on birth timing via processes
of the third kind mentioned above. Our estimates suggest that these other paths through
which rising education may have independently influenced birth timing account for,
at most, between 26 and 43 percent of the change in age at first birth over the period
(i.e. the complement of the overall structural effect). However, it could reach those
levels only if educational expansion were responsible for 100 percent of the upward
shift in mean ages at first birth.
A first route through which education affects birth timing is via time spent in full‐time
education. We noted earlier that micro‐level analyses using time‐varying covariates
have consistently found that first birth rates are substantially lower during periods
of enrollment. Economists use the term “incarceration effect” to describe this (see
e.g. Black et al. 2008; Monstad et al. 2008), but the term seems inaccurate. Young
people are no more incarcerated as full‐time students than they are as full‐time employees.
Instead, the enrollment effect is likely to operate through a mix of social and interpersonal
processes including social and economic dependency, expectations that the student
role is a transitional pre‐adult status, societal and peer‐group expectations regarding
the normal sequencing of adult transitions, and the age stratification and time‐criticality
of educational participation (Marini 1985; Hogan and Astone 1986; Liefbroer and Billari
2010). The limited resources of young people in education may also have an effect,
as suggested by Spéder and Bartus (2017). That educational participation inhibits
not only birth rates but also conception rates is suggested by Geruso et al. (2011),
who find that the raising of the school‐leaving age in the UK in 1974 caused a decline
in teenage birth rates but did not result in increased abortion rates at the ages
affected.
The second causal pathway is through the impact of age at leaving full‐time education
on the scheduling of adult transitions (see especially Oppenheimer 1994; Oppenheimer
et al. 1997; Robert‐Bobée and Mazuy 2005; Bergouignan 2006). On leaving education,
young people begin the process of finding employment, becoming financially independent,
accumulating experience, skills, and earning power, forming their own households,
finding a partner, and eventually starting a family. The later the age at leaving
education, the later this series of transitions begins; and the shorter the time out
of education, the less advanced individuals are in these transitions, including the
transition to parenthood. Figure 2 showed that first birth rates are strongly associated
with duration since leaving education, rising initially and then falling back. We
show elsewhere (Beaujouan et al. 2014b) that this duration outperforms age as a predictor
of first birth rates and makes a significant improvement to statistical models including
age terms only.
The findings of Skirbekk et al. (2004) reveal the impact of duration since leaving
education on first birth timing. They exploited a natural experiment created by Sweden's
primary school enrollment laws. These mean that, on average, people with December
birthdates are 11 months younger at completing their education, whether compulsory
or post‐compulsory, than those born a month later in January. As noted above, they
are also an average of 4.9 months younger on becoming a parent than those born the
following month.11 Skirbekk et al. suggest that the mechanism is what they term “social
age,” defined as the average age of a person's school cohort. However, their results
suggest that the link between school‐leaving age and first birth timing is better
explained via duration since leaving education than by social age. They find that
those born late in the year give birth to their first child at a younger age than
those born early in the year. This can be explained via the duration effect as follows.
All members of a calendar‐year cohort enter school and reach school‐leaving age at
the same time. Because of this, at any given duration since leaving, those born in
December will be younger than those born in January of the same calendar year. As
a result, they are subject to any given duration‐specific first birth rates at a younger
age than those born early in the year; this overrides the difference in age and results
in a younger age at first birth for those born late rather than early in the year.
A similar explanation, via duration effects rather than social age, can account for
the detailed pattern of differentials by birth month in age‐specific birth rates at
younger and at older ages reported in that study (Skirbekk et al. 2004: 557).
Concluding comments
While our analysis has been confined to three European countries, there are ample
reasons for thinking that a broadly similar picture holds in other developed societies.
We saw in Figure 1 that the trend to later first birth was common to a wide range
of European and other developed countries. A steady and continuing increase in educational
participation has also been taking place throughout the developed world. In view both
of the sizable causal effects of education on fertility timing reported above, and
of the range of countries in which such effects have been found, it would be surprising
if our findings did not hold much more widely. Large datasets that combine precise,
long‐run information at the individual level on both the ages at participation in
and leaving education and the age at first birth are, however, scarce, and so confirmation
of generalization to other contexts is not yet possible.
A final question is what accounts for the rise in educational participation. Change
in statutory compulsory schooling has played some role. Since the 1960s most European
countries have enacted legislation to increase the number of years of compulsory schooling
(Garrouste 2010: Figure 3.1; Murtin and Viarengo 2011). During the decades covered
by this study, European governments have continued to extend the compulsory requirement
for education and training (Eurydice/Eurostat 2012). In Belgium, the minimum age at
school leaving was extended from 14 to 18 years in 1983, a reform which, however,
followed a gradual development already taking place. Minimum school‐leaving age was
extended from 14 to 16 years in France in 1967 and from 15 to 16 years in the UK in
1973; compulsory participation to the age of 18 was introduced in the UK in 2013.
The actual change in enrollment over the period covered in this article is larger
than the change induced by compulsory schooling legislation. There is evidence that
compulsory schooling laws have the effect of increasing participation even beyond
the statutory provisions (Oreopolous 2009). Education expansion over the period analyzed
here has occurred in the context of concerted government effort to broaden access
and increase participation (Green 1999; Smith and Bocock 1999; Hansen and Vignoles
2005). Policy initiatives, in addition, appear to have had some success (Machin and
Vignoles 2006; Braga et al. 2013). Policy is driven, in turn, by changes in the labor
market, with the decline in unskilled jobs and the increasing demand for an educated
and skilled work force. Thus, rising education is propelled in part by macro‐economic
forces that not only influence governments to act but also motivate individual decisions
to participate. We conclude that both macro‐economic factors and policy interventions
that have promoted the substantial rises in educational enrollment are major contributors
to the fertility postponement seen across this period.
Supporting information
Supporting Information
Click here for additional data file.