Dr Bruce Bradbury, Social Policy Research Centre, University of New South Wales, Sydney NSW 2052, phone 02 9385 7814, email firstname.lastname@example.org
The research reported in this paper was completed under FaHCSIA’s Social Policy Research Services Agreement (2005–2009) with the Social Policy Research Centre. The opinions, comments and/or analysis expressed in this document are those of the author and do not necessarily represent the views of the Minister for Families, Housing, Community Services and Indigenous Affairs or the Australian Government Department of Families, Housing, Community Services and Indigenous Affairs, and cannot be taken in any way as expressions of Government policy.
This paper uses unit record data from the Longitudinal Study of Australian Children (LSAC) Survey. The LSAC Project was initiated and is funded by the Australian government Department of Families, Housing, Community Services and Indigenous Affairs (FaHCSIA) and is managed by the Australian Institute of Family Studies (AIFS). The findings and views reported in this paper, however, are those of the author and should not be attributed to either FaHCSIA or AIFS.
This paper uses unit record data from the Household, Income and Labour Dynamics in Australia (HILDA) Survey. The HILDA Project was initiated and is funded by the Australian Government Department of Families, Community Services and Indigenous Affairs (FaCSIA) and is managed by the Melbourne Institute of Applied Economic and Social Research (MIAESR). The findings and views reported in this paper, however, are those of the author and should not be attributed to either FaHCSIA or the MIAESR.
Bradbury, B. (2011), Young Motherhood and Child Outcomes, SPRC Report 1/11, prepared for the Australian Government Department of Families, Housing, Community Services and Indigenous Affairs, submitted November 2007, published January 2011.
List of Tables 6
List of Figures 6
Executive Summary 7
1.The Longitudinal Study of Australian Children (LSAC) 21
List of Tables
List of Figures
There is considerable evidence that childbearing at a young age is associated with poorer outcomes for both mother and child. However, international research suggests that much of this association is not causal – children born to young mothers might still have had poor outcomes even if their mother had delayed their childbearing.
Data from the Longitudinal Study of Australian Children (LSAC) shows that children aged 4-5 years whose mothers were under 25 when they were born have distinctly lower levels of functioning than those with older mothers.
Census data on 16-18 year olds who are still living with their mother show that this disadvantage carries through to education and labour market outcomes. Those born when their mother was in her teens are much less likely to be still in school.
We cannot, however, assume that all these differences are directly caused by being born to a young parent. It is quite possible that such associations could arise because of the different characteristics of the mothers (and fathers) who have their children when young. Testing for a causal relationship is important in formulating policy responses to these issues. If young motherhood causes negative outcomes for either the mother or child then this provides support for policies to discourage early childbearing. If, on the other hand, this association arises because of underlying socio-economic disadvantage that influences both young fertility and later outcomes, then such fertility interventions will have no impact on later outcomes.
Evidence from other countries is mixed. A recent US survey concluded that being born to a teen mother does not have an impact upon the school test scores of offspring, but may have an effect on other behavioural outcomes. Recent UK research on the other hand, has found a significant negative impact on schooling and employment outcomes.
This report uses several different methods to test for the existence of a causal impact of mother’s age at birth in Australia. Outcomes for young children and for teenagers/young adults are considered.
Using the LSAC survey, it is found that children born to mothers aged 30 have about half a standard deviation higher learning outcome score than children born to mothers aged 18. For social-emotional outcomes the effect is larger, at 0.7 standard deviations.
The report then controls for background socio-economic characteristics using two different approaches. The first approach holds constant a range of conventional socio-economic background variables. The second approach also holds constant the mother’s age at the birth of her first child. The results from the two approaches are similar. The association with learning outcomes disappears, but the relationship with social-emotional outcomes persists. However, given that the latter outcome score is entirely parent-rated, this result could possibly be due to the different expectations of parents of different ages.
The Household Income and Labour Dynamics in Australia (HILDA) survey is then used to examine outcomes for teenagers and young adults. The HILDA data allows us to control for all fixed characteristics of families (even those unobserved) by comparing the outcomes of siblings.
We find no significant difference in sibling year 12 completion (controlling for a first-born child effect). A similar story applies to youths’ self-ratings of their educational performance and life satisfaction.
These conclusions rule out any large effects of mother’s birth age on these outcomes, but the sample size in HILDA is insufficient to rule out modest impacts of birth age. For example, we cannot reject the hypothesis that the impact on school completion of being born to an under 23 year-old mother is as large as the 13 percentage point difference for teenage mothers found in UK research.
The results of this research cannot thus be described as conclusive. There is some evidence of an impact of mother’s age at birth on social/emotional outcomes of young children, but this could be due to parental expectations at different ages. For teenage outcomes, we cannot find any impact when comparing siblings, but larger samples are needed in order to rule out effects such as those found in some other studies in other countries.
There is considerable evidence that childbearing at a young age is associated with poorer socio-economic outcomes for both mother and child. The correlation between young childbearing and mother’s characteristics has recently been documented in the SPRS Young Mothers project. However, international research suggests that much of this association is not causal – the mothers might still have had poor educational and labour market outcomes even if they had delayed their childbearing. Similar conclusions for Australia were found in the SPRS project TheCausal Impact of Young Motherhood, though there is some evidence that young childbearing might reduce partnering opportunities for mothers.
Why might being born to a younger mother have adverse consequences for children? In summarising the literature on this, Levine et al (2005), identify a number of potential influences. In modern societies, having children when young may have a serious impact on the mother’s human capital development. Educational attainment might be reduced and entry into rewarding labour market careers disrupted. In turn, this might mean fewer economic resources and skills available to be transferred to the child. In addition, younger mothers may have less personal social development and hence have poorer parenting skills than more mature mothers. Related to this, they are less likely to marry the child’s father – which in turn reduces the economic resources of the family and reduces the likelihood of stable relationships between the child and father-figures.
However, it is difficult to find clear evidence that these hypothesised patterns actually have an impact on children. In their survey of observational studies on the parenting knowledge and behaviours of young mothers, Geronimus et al (1994) find conflicting evidence. Some studies find young mothers to be less sensitive and responsive, more likely to use restriction and punishment and to have less knowledge about parenting and child development. However other studies find no differences by age in these areas.
Indeed, it is likely that much of the association between young motherhood and poor child outcomes reflects the role of causal variables other than the fact of young motherhood itself. There is extensive evidence that women from socio-economically disadvantaged backgrounds are more likely to become young mothers (eg Stewart, 2003). If their children have poorer outcomes, this might be due to the impact of these background factors, rather than to young motherhood per se.
Identifying the causal relationships that underlie these observed associations is important in formulating policy responses to these issues. If being a young mother causes negative outcomes for either the mother or the child then this provides support for policies to discourage early childbearing. If, on the other hand, this association arises because of underlying disadvantages that influence both young fertility and later outcomes, then such fertility interventions will have no impact on later outcomes. In this case, being born to a young mother should be seen as an indicator of other disadvantages and possibly be used as an instrument to target assistance.
To statistically identify the causal impact of young motherhood on child outcomes, we need to compare outcomes of children born to young mothers with children who are identical in all other relevant respects, but who were born when their mother was older. The awkwardness of this thought experiment means that valid empirical tests are difficult to formulate. Nonetheless, there are several approximations to this which are considered in this report.
The report begins, however, with an examination of Australian evidence on the strength of the association between mother’s age at birth and child outcomes. We consider evidence both for young children (4-5 years old) and for teenagers. In both cases, children born to younger mothers have significantly poorer outcomes.
As noted above, however, it is quite possible that such associations could arise because of the different characteristics of the mothers (and fathers) who have their children when young. In Section we review the different methods that have been used by researchers to identify the causal impacts of mother’s age at birth. These include:
Controlling for observed characteristics (but many characteristics are not observed).
Using sibling difference (fixed effect) models to control for unobserved characteristics that are constant across siblings. (However, some characteristics, most notably birth-order, cannot be the same for all siblings).
Comparing children born to young mothers with children born to older mothers who had a miscarriage when young (but sample sizes are typically small).
Relying on the latter two, more rigorous, methods, a recent US survey concludes that teen parenthood does not have an impact upon the test scores of offspring, but may have an effect on other behavioural outcomes. Recent UK research on the other hand, has found an impact on schooling and employment outcomes.
In Section we examine these associations for pre-school children, using data from the first wave of the LSAC study. For both learning and social-emotional outcomes, children born to young mothers have poorer outcomes. When we control for family characteristics (including the mother’s age at the birth of her first child), the learning domain association disappears, but the social-emotional domain relationship persists (though it is smaller).
Section then examines outcomes among teenagers in the HILDA survey data. Here we compare siblings born when their mothers were of different ages, and examine whether they have completed year 12, how they rate their schooling performance, and how they rate their satisfaction with different areas of their lives. We find no evidence that the siblings born when their mother was younger have poorer outcomes. However the sample size in the HILDA survey is still relatively small. Section concludes.
In Australia there are around 11,000 children born to teenage mothers every year, and 36,000 born to those aged 20-24 (Bradbury, 2006a). This comprises around 4 and 15 per cent of all births respectively. In the late 1990s, the Australian incidence of teenage motherhood was the lowest of the English-speaking countries, but higher than in most other OECD countries (e.g. Germany, France, Sweden, Japan) (UNICEF, 2001).
There is strong Australian evidence of an association between a mother’s age at birth and her later socio-economic outcomes. Young mothers live in more disadvantaged regions, have lower levels of education, and by the time they are in their early-30s they have lower incomes and are less likely to be purchasing their own home (Bradbury, 2006a). This concentration of disadvantage has increased over the last 20 years. Though there is not as much research on the outcomes of the children born to young mothers in Australia, we would expect many of these forms of disadvantage to also apply to them.
Figure Average child functioning scores of 4-5 year-olds by mother’s age at birth
Source: LSAC. See Section for details. Unweighted data
Error: Reference source not found, from the first wave of the Longitudinal Study of Australian Children (LSAC) survey, shows how the average child functioning score for children aged 4-5 increases with the mother’s age at birth up until she is in her late 20s. This score encompasses the survey estimates of physical functioning, and social/emotional and intellectual capacities. The LSAC data are discussed in more detail in Section .
This relationship with mother’s age at birth is quite dramatic. For children born when their mother was aged 27 or older, outcomes are on average, much the same. For children born to mothers aged under 23, child functioning is almost a half of a standard deviation lower.1 The average outcomes of children born to teenage mothers decline even further, but because of the small sample size, these are not very accurately estimated in the survey.
This threshold of 23 years has not historically been considered young to be a mother, but it is now relatively unusual. In the LSAC data, only 12 per cent of the 4-5 year-old children were born when their mother was 23 or younger.2
As noted in the introduction, patterns such as those shown in Error: Reference source not found might arise from the causal impact of being born to a young mother, or they might reflect an association between the background characteristics of those women who have children when young and the outcomes of their children. We don’t know whether these children would have had better outcomes if their mother had delayed their birth.
Table Family status of youth aged 16-18 by their (apparent) mother’s age at birth
Note: The table population is youth aged 16, 17 or 18, living in a private dwelling on Census night.
(a) Households where the youth's natural or adopted mother was not in the household on Census night. NB the first two cells ‘both parents’ and ‘step-father’ appear to be a coding error in the Census data. This is still being investigated.
(b) Cells with less than 10 cases in the 1% Sample File. Blank cells have zero cases.
The sample size is the population (N) divided by 100.
Source: Calculated from the ABS 2001 Census Household Sample File (expanded version).
There are also strong associations between being born to a young mother and later life outcomes. Error: Reference source not found shows the family status of youth aged 16 to 18 in 2001. This table is calculated from the information collected in the 2001 ABS Census on the family relationships of people living in the same household on Census night. As such, there are some limitations in the mapping of family structure. The publicly available data does not distinguish natural from adoptive mothers and so the latter are treated as the natural mother for the calculation of the mother’s age at birth. About 21 per cent of youth aged 16-18 were not living with their natural or adoptive mother on Census night (156,700 youth). This includes those living away from their mother as well as those temporarily absent on census night (the latter are included among the ‘other’ family type in the table).
Among those still living with their mother, family structure varies greatly according to their mother’s age when they were born. About 80 per cent of those born when their mother was aged 30 or older are living with both their parents. Among those born when their mother was a teenager, only around half are with both parents. For this youngest-mother group, about a third are living with their mother only (i.e. as the child of a lone parent).
Though we again don’t know whether all of this association between young motherhood and later family characteristics is causal, there is evidence that much of it is. In earlier work (Bradbury, 2006b) I compare women who were young mothers with women who were pregnant when young, but had a miscarriage. Controlling for the attenuating impact of abortions, I find that having a child when young leads to a 24 percentage point reduction in the probability of being legally married at around age 30 (though there is no causal impact on being partnered per se). Young partnerships (legal or defacto) are generally less stable (Bradbury and Norris, 2005) and so it is quite plausible that young mothers will be less likely to be living with the father of their child by the time their child is a teenager.
Turning to more direct indicators of children’s outcomes, Error: Reference source not found shows the relationship between the education and employment status of youth aged 16 to 18, and their (apparent) mother’s age when they were born. In this age range, full-time study or employment are likely to lead to the most favourable adult outcomes (particularly the former), with those youth who are neither employed nor studying likely to have the poorest outcomes.
The group with the poorest outcomes are those not living with their (apparent) mother on Census night, with 19 per cent not engaged in employment or education. As noted above, this group includes those who have moved out of home, as well as those who were temporarily in a different household from their mother on Census night.
Among the remainder still living with their mother, outcomes improve with the increase in the age of the mother at birth. The proportion in full-time study increases from 60 to 80 percent as mother’s age increases, while the proportion not in study or employment falls from 15 to 5 per cent. As for the outcomes for young children, the main differences are found among those with mothers in the younger age groups, with only small differences among those with mothers aged above 25 at birth.
Table Education and employment status of youth aged 16-18 by their (apparent) mother’s age at birth
Note: The categorisation into education/employment category is hierarchical, with people classified into the first category into which they fall. For example, a person studying full time and also working part time would be placed into the ‘Full-time study’ category. The ‘not stated’ group are those youth who did not fall into any of the first three groups, and who did not answer either the education or employment status questions. Mother’s age at birth is calculated by identifying the woman who was the head of the family in which the youth was recorded as a child. In some cases this may not be the youth’s biological mother (e.g. an adoptive or step-mother). A small number of cases with unrealistic mother ages (which can arise if the father re-partners with a younger wife) are excluded from the table. The sample size is the population divided by 100.
Source: Calculated from the ABS 2001 Census Household Sample File (expanded version).
Identifying the impact of maternal age on child outcomes
However, as already noted, these associations do not necessary imply causation. Women who have poor educational prospects are more likely to become young mothers, and there is likely to be some degree of correlation between the generations in their educational attainments even in the absence of an age-at-birth effect on children. Similarly, mothers who themselves have emotional or intellectual characteristics that would score poorly on the indicators measured in the LSAC survey, may pass some of these characteristics on to their children (via genetic and/or behavioural transmission). These same characteristics may imply poor prospects in the education system and the labour market and so make parenting at a young age relatively more attractive. In addition, these prospective parents may have fewer opportunities to control their fertility (e.g. living in regions with fewer services). Both sets of factors will mean that disadvantaged children will be more likely to be found in families with younger parents. However, these children might still have had the same outcomes even if their parents had delayed their childrearing.
To statistically identify any direct impact of mothers’ (or fathers’) ages on child outcomes, it is necessary to control for these background characteristics in some way. Unfortunately, many of these characteristics (such as personality traits) are difficult to measure. Several methods have been used in previous research to address this. These include the following approaches (the terms in italics are used to refer to these approaches in the following text):
Using statistical techniques to control for the observed differences between children born to young and those born to older mothers. The main limitation of this is that many key characteristics are typically not easily observed and measured.
Using fixed effect models to control for family background. These include comparisons of siblings and comparisons of cousins. Comparing cousins controls for the characteristics of the mother’s family background, but does not control for the specific characteristics of each mother. The more high-achieving sister might be both less likely to have a child early and also more likely to have high-achieving children. Comparing siblings is better in that it controls for any constant characteristics of the mother, but is confounded with the effects of birth order on outcomes.
Comparing children born to young mothers against children born to older mothers who had been pregnant but had a miscarriage when they were young. This is probably the most robust approach, but few datasets have the required information on both miscarriages and children’s outcomes, and those that do have relatively small sample sizes.
Typically, studies that control for observed differences find that the amount of association between maternal age and child outcome measures diminishes, but does not disappear, when parental characteristics such as socioeconomic status are controlled for (Geronimus et al., 1994). The other methods typically find much smaller effects of being born to a young mother, though they are more demanding of the data and thus are less likely to identify any true effect if that effect is relatively small in magnitude.3
The recent paper by Levine et al (2005) compares results using several different methods, and also examines a range of outcomes for adolescents and young adults. They use data from the US National Longitudinal Study of Children and Youth (NLSY79) and the associated survey of the children of these youth. This dataset has been widely used for the study of intergenerational linkages in characteristics and outcomes. The outcome measures examined include grade repetition and test-outcome scores in mathematics and reading, together with behavioural outcomes such as sexual behaviour below age 16, marijuana consumption, fighting and truancy.
They use all three of the observed differences, fixed effect (pooling siblings and cousins together) and miscarriage methods. For learning outcomes, they find consistent outcomes across both the fixed effect and miscarriage models – having a teenage mother has no causal impact. However, they find mixed results for some of the behavioural outcomes. Grade repetition, truancy and early sex are associated with being born to a younger mother in the observed differences and fixed effect models, but have an opposite but insignificant effect in the miscarriage model. Though the miscarriage model is theoretically the best way to hold constant many of the unobserved background characteristics, the small number of miscarriages in the sample means that they cannot conclusively state that being born to a teen parent has no causal impact on these behaviours.
After comparing their work with that of other researchers, they conclude that teen parenthood does not have an effect on the learning outcomes of the children: “The totality of findings should settle the question of whether early childbearing affects test scores of offspring” (p120). We should temper their firm conclusion with the knowledge that all these studies were based on the same dataset (the children of the NLSY79 study).
Indeed, others have drawn different conclusions. Using data from the British Household Panel Study (BHPS), Francesconi (2007) examines patterns using both the observed differences and sibling comparison (fixed effect) methods.4 Focussing on the latter method, he finds several adverse effects of being born to a mother who is a teenager (compared with the outcomes of the person’s sibling born when the mother was older). The children of teenage mothers are 13 percentage points less likely to achieve an A-level or higher qualification, are more likely to be economically inactive and have lower income and, if female, more likely to be a teenage mother herself. No impact on smoking and psychological distress is found. Evidence is presented that suggests that the unfavourable outcomes act via their relationship with childhood family structure (i.e. children born to young mothers are more likely to live in lone-parent families). The adverse effects of being born to a young mother also apply to mothers in their early 20s (though are attenuated).
In this study we use two Australian longitudinal datasets to investigate these issues. Neither is as comprehensive as the NLSY, though one has similar data to the BHPS.
The first wave of the Longitudinal Study of Australian Children (LSAC) is used to examine developmental outcomes among 4-5 year-old children. Error: Reference source not found provided a summary of the relationship between child outcomes and mother’s age at birth. In Section this data is examined more closely. Different outcomes are separately examined, we control for socio-economic background, and we also control for the mother’s age at the birth of her first child. The LSAC survey studies one child from each family, who could be a first or a subsequent child (depending upon who is aged 4-5 at the time of sample selection). Controlling for the age of the first child is a very comprehensive way of controlling for any impact of the mother’s background on fertility patterns.
With the Household Income and Labour Dynamics in Australia (HILDA) survey, it is possible to compare siblings, though the sample size of matched siblings is (currently) smaller than the BHPS. This survey does not collect detailed information on the outcomes for young children, but it can be used to look at outcomes for older children. The key outcome examined is attainment of year 12, though we also consider some more subjective indicators of teenagers’ well-being.
Since both these analyses rely heavily on the comparison of siblings (or controlling for sibling ages in the LSAC data), the method of sibling comparisons is described in more detail below.
Sibling fixed-effect models
The essence of the sibling fixed-effect model is to compare the difference between two siblings on an outcome measure against the difference in their environment. The latter might include factors such as mother’s age at birth, family structure when young, etc. This differencing holds all the unmeasured fixed characteristics of families constant, and hence controls for any fixed confounding causal effects such as family background or mother’s personal characteristics.
The key assumption of the method is that any unobserved confounding effects are the same for both siblings (i.e. fixed). It is also necessary to make assumptions about the functional form of the impact of these factors so that they can be ‘differenced out’.
More specifically, let yif represent the outcome for the ith child in the fth family. This is assumed to be a linear function of the age of their mother when they were born, aif , unobserved family characteristics, uf, which are assumed to be the same for all siblings, a vector of other observed characteristics Xif, and a random error eif.5
If we cannot observe the background family characteristics uf we are forced to estimate the relationship
If the omitted background characteristics, uf, are correlated both with mother’s age at childbirth, a, and with the outcome variable, then estimates of b will be biased estimates of the impact of mother’s age on outcomes.
This problem can be addressed by using sibling differences. Starting from equation Error: Reference source not found, for a two-child family we can write
That is, we estimate the relationship between the sibling-outcome differences and the difference in mother’s age at birth for the two siblings. Because both siblings have the same family background characteristics, these drop out of the estimation. (Similarly any components of X that are same for the two siblings are set to zero).
In this report, siblings are defined as children with the same biological mother, and so is equal to the age difference between the siblings. More generally, if we wish to test for the impact of being born to a mother in a particular age range, then we can replace the continuous age variable with one or more dummy variables indicating the age range. These can be differenced across the siblings in the same way as the continuous variables.
Under either approach, this sibling difference estimation amounts to testing whether (some) younger siblings have better outcomes than older siblings. Stated this way, one estimation challenge of this sibling difference approach is apparent. Though the method effectively controls for unobserved characteristics that are constant over time, (non-twin) siblings all differ in one important aspect in addition to their age – the presence or absence of siblings. In a two-child family, the sibling born to the mother when she was youngest is the first-born sibling and will spend the first year(s) of their life as an only child, i.e. without any siblings. Similarly, the sibling born to the older mother is always a second child (in the absence of child mortality).
Other unobserved differences between the siblings can also bias the results of sibling difference estimates. Such differences might include changes to the social environment over time (e.g. government policy, social norms, etc) and changes in the family environment (e.g. different fathers, etc). If these are systematically different across the whole sample, and are also associated with child outcomes, this will bias the results. For example, if schooling policies addressed at retaining children to year 12 became more prevalent over time, then we might find that the younger sibling has a better schooling outcome. Such effects, however, are probably not too much of a problem as there is an average of only three years between the members of the sibling pairs.6
In addition, causal links between the siblings would also invalidate the fixed effect assumption. For example, if a frail first-born child discourages a mother from having a second child, then the second-born children in two-child families will comprise only those children with non-frail siblings. The first-born child, on the other hand, will have siblings with the full range of frailty levels. This unobserved difference between the siblings (selection bias) would thus lead to the erroneous conclusion that being born to a younger mother leads to better outcomes.
The most obvious confounding factor, however, remains the birth-order effect. This is different by definition between the siblings and it could lead to different outcomes. To see whether it is possible to separate age at birth from birth-order effects we need to more closely look at the mechanisms posited in the literature. Drawing on the survey by Booth and Kee (2005), the key hypotheses discussed in the birth-order literature are summarised in Error: Reference source not found. They are categorised here according to whether they imply better or worse outcomes for first-born children, and also according to whether we might expect to find a stronger effect among younger or older mothers.
Table Hypotheses of the impact of birth order and age difference on child outcomes
Greater family expenditure needs in larger families
Parents are more experienced when raising later children
equal or younger?
Additional caring available from older siblings
Greater responsibility given to older siblings
Older siblings leaving school early to provide for younger siblings
These hypotheses are separated here into three groups: parental-age effects, birth-order effects and sibling-interaction effects. The focus of this report is on parental-age effects, but we also need to consider the impacts of the associated birth-order and sibling-interaction effects.
In the literature on young motherhood, it is speculated that young mothers (and also fathers) may be less competent in a range of dimensions than older mothers. This may be the case even in the absence of selection effects. That is, parents may become generally more competent as they age, implying a higher standard of care for the later-born siblings. If there is such an age effect, it is plausible that there will be decreasing returns to age. A mother aged 21 might be substantially more competent than when she was 16, but an additional 5 years might make little difference. This is certainly how the literature in this area is framed. There is a particular concern about very young mothers, but less concern about differences between older mothers. Hence, the last column of this table describes this hypothesised impact as being stronger for younger mothers.
Similarly, family income-earning potential (and economic resources more generally) also increases greatly during the late teens and early 20s but usually less thereafter.
At the other end of the age scale (over age 35) biological disadvantages with giving birth start to appear. Hence there might be poorer outcomes among the later-born siblings if they are born late enough.
Confounding these effects of maternal age, however, are birth-order and sibling-interaction effects. Earlier-born children might do better because their mothers are healthier, and/or they get more parental attention and/or a greater share of the household’s economic resources (i.e. they spend more time in a smaller family). On the other hand, the later-born children might do better because the parents have gained more child-rearing experience. The literature generally concludes that first-born children do fare better on average than later-born, but that there is little difference between the children after the first-born.
For the most part, there is no particular reason to think that these birth-order effects should be different among families where child-rearing starts earlier or later. However, it is possible that the parental-experience effect is less for older parents if they have been able to gain experience and caring skills elsewhere.
Similarly, there might be sibling-interaction effects, with the older siblings either benefiting or being disadvantaged depending on the patterns of additional responsibility placed on them. There is little reason to believe that these patterns will vary with the age of parents.
These different patterns of effects suggest some possible identification strategies to separate out birth-order from the effects of mother’s age at birth.
One strategy is to control for being a first-born child while examining the impact of mother’s age. If we assume that 1) the birth-order effect applies only to the first child (this is consistent with the literature); and 2) that the first-born effect is the same for both younger and older families (e.g. ignoring any possible differences due to the parental-experience effect); and 3) that the additive structure of the model is correct (likely to be at least approximately correct), then this provides a way of estimating the independent impact of being born to a young mother. This approach is used in Section below.7
A related strategy is to draw on the theoretical assumption that the parental-age effects are likely to be stronger for pairs of siblings born to younger mothers (again ignoring the parental-experience patterns).8 Hence if the observed combined effect is concave downwards across mother’s age, then this will provide some support for an effect on child outcomes of mother’s age at birth.
Pre-school outcomes and parental age: evidence from the Longitudinal Study of Australian Children