There are two main hypotheses that I test in the empirical analysis:
a) Is there a causal effect of family size on child psychological well-being and cognitive ability?
b) Is there an association between a higher birth order and child psychological well- being and cognitive ability?
In sections 5.3.1-5.3.3 below, I outline the theoretical reasoning behind both hypotheses. 5.3.1 Family size
A number of theoretical models from the economic, psychological and sociological literatures have considered the relationship between family size and child outcomes. The most prominent of these frameworks in the economic literature is the Quantity vs Quality model (QQ model) of fertility, presented by Becker and Lewis (1974). This model treats children as analogous to consumer goods, with parents deriving utility from both the quantity and ‘quality’ of children, as well as the consumption of other commodities. Given that there are fixed time and budget constraints and parents are utility maximising, this model showed that there may be a trade-off between the quantity and perceived ‘quality’ of a child, as additional children increase demands for both financial resources and time inputs of the parents. Therefore, the QQ model predicts that children born in larger families may be hindered, as they have to share resources and time with their parents and their siblings. Becker and Tomes (1976) extended the Becker and Lewis (1974) model to integrate social interactions, in order to analyse the robustness of the QQ model to several external factors. For instance, given that an increase in parental income will lead to a large increase in
112
children would have to come from an increase in these expenditures, due to the fact that child endowments are assumed to be fixed. This in turn will cause the demand for children to be reduced. This implies, for example, that the QQ trade-off may be more pronounced amongst lower socioeconomic groups than higher socioeconomic groups.
Alongside this influential economic model, the Confluence Model (Zajonc 1976) and Resource Dilution Model (RD model) (Blake 1981) stem from the psychological and
sociological literatures respectively. Rather than economic resources, the Confluence Model argues that the ability of a child is dependent on the average intelligence of the household. Given that the arrival of a new child (initially with no intellectual skill) will decrease the average intellectual level in the family, large families will provide a more immature environment, which may negatively influence the child’s level of intelligence. Rather than average family intelligence, the RD model relates child ability to the home environment created by the parents, whether this being the quality of the learning environment, outside activities or personal attention. Given that additional siblings will reduce (or dilute) the proportion of resources received by any one child, this may impact the perceived ‘quality’ of the child.
Although the three theories discussed above dominate the theoretical literature regarding the impact of family size on child outcomes, a number of authors have instead argued that the relationship between family size and child outcomes may not be so clear cut. For instance, Velandia et al., (1978) and Page and Grandon (1979) have argued that the
empirical implications of the above theories may be better explained by a set of observable and unobservable household level ‘admixtures’ potentially relating to both family size and child outcomes, for example social class and ethnicity. Similarly, Rodgers et al., (2000) has theorised that family size has little causal effect on child outcomes, instead arguing that the observed family size differences may be in fact working through a non-behavioural
component of the model, this being the homogeneity of the intelligence within a family compared to between families. The authors point to the fact that most of evidence
supporting the QQ, confluence and RD models come from cross sectional data, which may include a number of biases, and advocate the use of within family data to investigate a within family problem.
113 5.3.2 Birth order
As well as family size, a number of theoretical models have also been presented to explain potential birth order differences in child outcomes. Ejrnaes and Portner (2004) have argued that these models can be divided into a number of categories, including financial and time constraints, household environment and biological effects.
From an economic perspective, a number of studies have argued that birth order effects may be explained by a variation of the human capital model, in which parents are faced with various financial and time constraints over the life course. For instance, Birdsall (1979) argued that given there is only a limited amount of time a parent can spend with their children, an eldest child will spend more time with the parents compared to later born children, particularly in the crucial early years of life. Building on this framework, Behrman et
al., (1982) and Behrman and Taubman (1986) have argued that the extent of these predicted
birth order effects may specifically depend on the preferences of the parents. If the parents are non-discriminatory between their different children, they will allocate the same amount of time to each of their children. However, as argued by Hertwig et al., (2002), dividing up resources equally amongst different children at each distinct time point may itself create inter-temporal inequities between different birth orders. If parents instead attempt to maximise overall achievement, and therefore their utility, they will put a higher level of resources on the more productive children. In this case, the addition of a higher quality child will exacerbate the problem of an extra child. Alternatively, if parents seek to ensure that all of their children have equal outcomes, they may divert a higher level of resources to less productive children to compensate for their lack of productivity.
The household environment explanation of birth order effects relates back to the confluence model (Zajonc 1976). Given that the model predicts that child ability may be determined by the intellectual environment the child grows up in, this implies that children further down the birth order are at a distinct disadvantage, as they will grow up in a lower intellectual environment compared to their older siblings. These effects may be particularly large for the last born child, as they do not have any younger siblings to help teach. However, this model also implies that such effects can be heavily mediated by larger spacing between births, and the relative intellectual ability of the child’s siblings.
114
The biological explanation of negative birth order effects relates to the impact of maternal depletion on birth endowments (Behrman and Taubman 1986). Given that later born children will by definition have older mothers, it is argued that this may advantage older children, as older mothers tend to have children of lower birth weight, are more likely to have children with birth defects and are more likely to have dizygotic multiple births, all of which are associated with a number of adverse child outcomes.
However, although the majority of the theoretical models have pointed to a negative relationship between later birth order and levels of child ability, parts of this theoretical literature have instead argued that there may be advantages of having a higher birth order. For instance, using a model relating to the intra-household allocation of resources in
conjunction with endogenous fertility, Ejrnaes and Portner (2004) have argued that parents may decide to stop having children when the genetic endowment of the last born child is higher than expected, and that therefore parents may in fact favour the last-born children. The authors also note that the expected compensatory behaviour between heterogeneous children may not be observed due to inequality-averse parents only having one child. Furthermore, a number of authors (Behrman et al., 1982; Behrman and Taubman 1986; Hertwig et al., 2002) have argued that in economic terms, having older parents may in fact be considered an advantage, as older parents may be more responsible and mature, and therefore may also be closer to reaching the peak of their earnings profile. Consequently, siblings further down the birth order may benefit from the increase in family income over time, as parents may be able to dedicate proportionally more financial resources on children lower down the birth order compared to their older siblings (Parish and Wills 1993). Finally, although several authors have argued that in biological terms, having higher maternal age at birth may be considered a hindrance to child development, other studies have argued that this increased maternal age may in fact be an advantage, mediated either through the mother’s womb becoming more effective at nurturing a foetus (Khong et al., 2003) or successive children being hypo-masculinized by maternal immunization to the H-Y antigen (Beer and Horn 2000).
115 5.3.3 Empirical implications
From the various theories considered in subsections 5.3.1 and 5.3.2, one can relate how both family size and birth order may be related to child outcomes such as psychological well- being and cognitive ability. For family size, the majority of the theoretical frameworks, including the influential QQ model, predict that having a larger family size will have a significant negative effect on child outcomes, due to the dilution of parental resources between the increased numbers of siblings. For birth order, the majority of the theoretical frameworks presented argue that being a later born child will also have a significant negative effect on child outcomes, through the differing time spent with parents relative to other siblings, the household environment and biological effects. However, the theoretical literature is not universally in favour of negative effects of both family size and birth order, and the various models imply that that there are a number of other factors that may impact the strength and direction of the predicted relationships, such as socioeconomic factors and the home environment. Therefore, in the empirical analysis, it is important to account for these potentially confounding observable factors, in an attempt to isolate the specific effects of both family size and birth order on child outcomes16.
Given the arguments of Becker and Tomes (1976) regarding the robustness of the QQ model to external factors such as household income, it is clear that it is important to control for a wide range of socioeconomic factors that may influence the level of resources invested in the child, such as levels of household income and parental occupation, as it is likely that the trade-off will significantly differ across socioeconomic groups. Similarly, the confluence model (Zajonc 1976) implies that the average level of household intelligence may influence child ability. Although data limitations meant that I was unable to control for the intelligence of the other siblings present in the household or paternal education, I was able to control for the highest educational attainment of the mother. The inclusion of this education variable may also be able to help control for intergenerational transfer of ability, given the strong predicted relationship between maternal education and child outcomes noted by studies such as Carnerio et al., (2013).
The confluence model also implies that the spacing between siblings may impact the intellectual environment children grow up in, and therefore may influence child outcomes.
16 It should be noted that an additional requirement in the 2SLS models is that controlling variables should be
116
To help control for this, I included a measure of the average birth spacing between siblings for those children with siblings. As predicted by the RD model, another factor that may influence levels of child ability is the home learning environment. In order to control for the differing home environments that children may encounter, I included three variables related to the home learning environment from the MCS: the amount of time parents take reading to their children, how often the child draws and paints at home and the number of trips to the library.
Parish and Wills (1993) have argued that there may also be significant life-cycle effects which could impact the outcomes of children of different birth orders in a variety of ways.
Therefore, I included a number of factors which are expected to vary with maternal age, such as employment status, the birth weight of the child and how long the child was breastfed.
Although controlling for a wide variety of socioeconomic and household characteristics may be able to account for a significant amount of the potential confounding and mediating characteristics predicted by the theoretical models, the ‘admixture’ model favoured by Velandia and Page (1978) and Rodgers et al., (2000) still predicts that the effect of family size from such models will be biased, due to the importance of both unobservable between- family processes which may be related to child outcomes, and the confounding effects of birth order. In empirical analysis, I attempted to account for this possibility through the use of 2SLS models, which seek to isolate a causal effect of the endogenous family size variable by utilising plausibly exogenous variation in family size.
For birth order, several studies have argued that differences in child outcomes according to birth order may be driven by the inequitable distribution of resources within families (Rodgers et al., 2000) and that between household surveys may not be appropriate. Due to the nature of the dataset used, I was unable to account for birth order differences within individual households17. Therefore, my specific empirical strategy involved both estimating birth order differences within specified family sizes, and controlling for a wide range of potentially confounding characteristics. As a further step, my empirical strategy also involved the use of non-parametric NNM models, which do not impose a strict functional form and by
17 I am however comforted by the fact that the few studies in the literature that have used family fixed effects
models to control for within family variation have noted very little difference in the empirical estimates (Black
117
definition only consider pairwise comparisons within a region of common support, meaning that there is at least one match for each included observation (Cerulli 2015).