Due to the variety of household and longitudinal datasets available, the majority of the previous empirical work regarding the relationship between SES and child cognitive ability has been carried out in the UK and the USA. Although a significant empirical literature has investigated the relationship between measures of SES and other child outcomes, for example years of schooling (Duncan et al., 1998), completion of higher education (Carneiro and Heckman 2002) and health (Khanam et al., 2009), in the interests of space I focus
31
specifically on studies looking directly at the relationship between SES and child cognitive ability. The vast majority of studies which have examined the relationship between SES and child cognitive ability have found evidence of socioeconomic inequalities, despite differences in the setting, methodology and measurement of both cognitive ability and SES. A selection of the more notable studies are discussed below.
Duncan et al., (1994) used data from the USA based Panel Study of Income Dynamics (PSID) to estimate the impact that family income and persistent poverty status have on child IQ level, measured at age 5. Using linear regression analysis and controlling for a wide range of confounding characteristics including the home environment and neighbourhood factors, the results showed that an increase in income of $10,000 was associated with an increase in IQ at age 5 of 0.15 SD. Furthermore, the authors found that the impact of persistent poverty was roughly twice as large as the effect of transient poverty, and that the association was mediated by maternal depressive symptoms and the home learning environment. Utilising a sample of children (N=6864) from the USA based National Longitudinal Survey of Youth (NLSY), Blau (1999) estimated the effect of family income on a range of child
outcomes, including cognitive and behavioural development. Utilising a variety of panel data models in order to partially control for the endogeneity of income, results showed both current income and permanent income to be associated with child cognitive ability. However, the magnitude of this effect was found to be relatively small when a number of controlling characteristics were included in the empirical specification, implying that a range of factors associated with both income and child cognitive ability may have explained a significant proportion of the correlation. A number of other empirical studies (for instance Parcel and Menaghan 1990; Hill and O’Neill 1994; Korenmann et al., 1995 and Smith et al., 1997) have also investigated the relationship using the NLSY but less sophisticated
econometric methods, with the findings mostly in line with those from Blau (1999). In a highly influential and UK based study, Feinstein (2003) used a sub-sample of children (N=1194) from the BCS to show the long shadow that parental SES (as measured by parental occupational classification) has on child development (as measured by the wide range of cognitive tests available in the BCS). Empirical results showed that children from lower socioeconomic backgrounds had lower cognitive scores in later childhood, even if they had high cognitive scores in early childhood, with children from higher socioeconomic
32
certain proportion of the disadvantages displayed in this study (in particular the phenomena of lower ability children from higher social classes overtaking high ability children from a lower social class at a very early age) may in fact be attributed to regression to the mean (Jerrim and Vignoles 2013), the significant socioeconomic inequalities in cognitive ability are still apparent.
In a rare cross country comparison, Aughinbaugh and Gittleman (2003) examined the effect of household income on levels of child development in sub-samples of children from both the NLSY (N=2380) and the NCDS (N=2080). Similar to the studies of Duncan et al., (1994) and Blau (1999), results across both cohort studies showed a remarkably similar statistically significant association between levels of income and child cognitive ability, despite
significant differences in factors such as health care provision, racial composition and educational institutions. However, this effect was found to be relatively small (a $10,000 increase in income associated with a 0.05-0.08 SD change in cognitive ability) compared to other family background variables such as the home learning environment.
Goodman and Gregg (2010) used a variety of British studies to analyse the gap between the rich and poor in terms of educational attainment, including the MCS, Avon Longitudinal Study of Parents and Children (ALSPAC) and the Longitudinal Survey of Youth (LSY). Using parental occupational classification as their measure of SES and a wide range of measures of cognitive ability, results showed those children from households in the lowest quintile of a combined measure of SES had cognitive scores 23% lower than those in the highest quintile at the age of 3, with this level of inequality rising to 27% at age 5. Further analysis showed that a significant proportion of the gap in test scores between the richest and the poorest children could be explained by parenting behaviours and the cognitive ability of the parent, implying that this may be a potential pathway through which socioeconomic inequalities may be reduced.
Unlike the vast majority of the UK based literature, Violato et al., (2010) focused on the relationship between parental income and cognitive ability using both cross sectional and panel data regression methods. Once more utilising the rich MCS data, empirical estimates from both random effects and fixed effects model specifications showed that although family income was significantly associated with measures of child cognitive ability at age 5 (a one unit increase in logged permanent income was associated with an 0.1 SD increase in cognitive ability), the magnitude and precision of this estimate significantly diminished whilst
33
controlling for a variety of other factors. The authors also acknowledged that family income was likely to be acting as a proxy for a broader range of socioeconomic factors, and
therefore may not have a strictly causal interpretation.
Rather than estimating a conditional association between a measure of SES and child
cognitive ability, Milligan and Stabile (2011) exploited exogenous changes in child benefits in Canada to estimate the causal impact of household income on child cognitive ability using IV methods. Utilising the National Longitudinal Survey of Canadian Youth (NLSCY), results showed a relatively small causal effect of income on both maths and reading test scores in the full sample (a $1000 increase in income corresponding to a 0.03-0.07 SD increase in cognitive ability), with these effects larger among boys and those from families with low levels of educational attainment.
Several other studies in this literature have also attempted to account for endogeneity and estimate a causal effect of income on child cognitive ability, with these studies in general generating mixed results. For instance, Loken (2010) used the 1970s Norwegian oil boom as an instrument to find no causal relationship between income and measures of child
cognitive ability, while Loken et al., (2012) used the same natural experiment to find a small positive causal effect of income at the lower end of the income distribution. Furthermore, Dahl and Locher (2012) used non-linear changes in Earned Income Tax Credit in the USA to show that a $10,000 increase in income increased standardised cognitive ability by between 2-3%. However, the conclusions from this study are disputed, as Lundstrom (2017) has shown that a coding error when calculating the income variable may in fact explain a significant proportion of the estimates.
Most recently, Dickerson and Popli (2016) used the MCS to identify the relationship between persistent poverty and various measures of child cognitive ability from ages 3-7. Using structural equation modelling (SEM) methods in order to identify both the direct and indirect effects of poverty on cognitive development, empirical estimates showed that children born into poverty have a significant disadvantage in terms of cognitive ability after controlling for various background characteristics and measures of parental investment. The authors further noted the potential important role of parenting skills and investment, and also showed that poverty crucially has a cumulative negative effect. However, as argued by the authors, disentangling the effect of income from other household factors and treating
34
the estimates as causal may be difficult, given that children in poor households often have young, less educated and single mothers.
Although not discussed in detail, there have also been several other important contributions to this empirical literature. For instance, Wolfe (1982), Brooks-Gunn and Duncan (1997), Klebanov et al., (1998), Duncan et al., (1998), Taylor et al., (2004) have analysed the
relationship between SES and child cognitive ability using US data, while McCulloch and Joshi (2001), Gregg et al., (2007), Barnes et al., (2010), Kiernan and Mensah (2009; 2011) and Schoon et al., (2012) have analysed the relationship using data from the UK. All of these studies showed a significant association between measures of SES and child cognitive ability. One common feature of this empirical literature is the use of purely regression based
methods, with very few empirical studies having utilised more sophisticated measures of socioeconomic inequality, such as the concentration index (CI). As Wagstaff et al., (1991) have argued, the CI can be regarded as one of the most appropriate empirical measures of socioeconomic inequality, as it reflects the experiences of the entire population, is sensitive to changes in the distribution of the population across socioeconomic groups and
summarises the extent of inequality in a single measure that can be compared across groups.
Only two other published empirical studies have used the CI in the context of non-health child outcomes. The first of these was Maika et al., (2013), who analysed the change in socioeconomic inequality in child cognitive ability between 2000 and 2007 in the Indonesian Family Life Survey. Empirical results showed that although the burden of poorer cognitive function was consistently higher among the disadvantaged, this level of disadvantage decreased over time. Decomposition analysis showed household income and parental education to be the largest contributing factors to the overall level of income related socioeconomic inequality.
The second study to use the CI in child non-health outcomes was that of Vallejo-Torres et al., (2014), who investigated income-related inequality in a measure of psychological wellbeing, along with several other health measures, in five years of pooled data from the Health Survey for England. The results showed a significant level of socioeconomic inequality in child psychological well-being, with these inequalities being larger than those found in late adolescence and also larger than several domains of physical and mental health.
35