Is it enough to be bright? Parental background, cognitive ability and educational attainment

Lazarsfeld (1939) suggested that social indicators can be interchanged, meaning that one could take one indicator of social stratification or another and the outcome would overall be the same. Thus: ‘for the measurement of economic influence, different indices of economic status are often interchangeable’ (p. 38). This thesis has been widely accepted (even if the propagators may never have heard of Lazarsfeld's suggestion). For example, medical researchers, looking for the social gradient on health, have used many indicators – the Registrar-General's social class; house ownership; car ownership to mention three – and typically found a gradient, which seems to support the thesis. Adherence to Lazarsfeld's thesis implies that various indicators of social position are about similarly good indicators of a latent stratification dimension. Epidemiologists in studying social differences in health may schematically be said to have put much effort and ingenuity in setting up their dependent variables, that is, various measures of health, but have many times been rather haphazard in their choice of independent variables. In recent years sociologists have questioned this practice, particularly pointing out that in order to get a better grip of the mechanisms by which social positions influence health and other outcomes, precise measures of social positions are needed (Chan and Goldthorpe 2004, 2007; Torssander and Erikson 2009, 2010; Buis 2013; Bukodi and Goldthorpe 2013)

One indicator or another can be used if the sole purpose is to show that there are differences between social positions. However, a better understanding of the stratification process, of how life chances are structured by people's social positions, cannot be reached unless the analysis is based on fundamental stratification dimensions. Against this background, the aim of this paper is to investigate how four dimensions of social background – parents’ education, social class, social status and earnings – are associated with educational attainment and to discuss mechanisms that possibly could relate the four background factors to educational attainment and particularly to investigate the role of cognitive ability in this respect. I first discuss the part played by cognitive ability in the intergenerational transmission of advantage and how best to account for the social background and then raise two questions to be addressed. The results are presented after the data and the analytical approach have been described. The paper ends with a discussion and a conclusion.

Ever since Burt (1959, 1961) suggested that social mobility is largely attributable to variation in intelligence, there has been an ongoing discussion of the role of cognitive ability in social mobility (cf Saunders 1997, 2002; Breen and Goldthorpe 1999, 2002; Nettle 2003; Deary et al.2005; von Stumm et al.2009; Johnson et al.2010). In this paper I will take another look at this issue or, more specifically, at the importance of cognitive ability for the association between social origin and educational attainment – if anything, cognitive ability can presumably be expected to be more important for this association than for that between social origin and social class. While most earlier contributions have used parental social class or some composite measure of social background to indicate social origin, I will look at how several background factors – parents’ education, social class, social status and earnings – influence children's educational attainment and to what extent these associations are transmitted via cognitive ability.

There clearly is an association between the social positions of parents and their children's cognitive ability as well as with the children's educational level. Likewise, it is clear that individuals’ cognitive ability has an effect on their educational attainment independently of the social background, for example, by increasing the probability of bright, working-class children attaining higher education and professional positions and, perhaps to a lesser extent, increasing the probability of less-bright upper middle class children to slide down the stratification ladder (cf Galindo-Rueda and Vignoles 2005). Several papers (Deary et al.2005; von Stumm et al.2009; Johnson et al.2010; Cheng and Furnham 2012; Sorjonen et al.2012) observe, through structural equation modelling, that latent or observed characteristics of mental ability as well as some measure of social background are associated with educational attainment. In this paper I intend to contribute to this research by presenting more evidence about the importance of cognitive ability for the association between social background and achieved education.

Cognitive ability seems to be a relatively more important factor for educational attainment than social background (von Stumm et al.2009; Cheng and Furnham 2012; Sorjonen et al.2012). Furthermore, cognitive ability seems to account for a large part of the association between the education of parents and children. Using a large Swedish data set, Mood et al. (2012) found that among men cognitive ability at the age of 18 accounts for 37% of the association between fathers’ and sons’ education. However, Deary et al. (2005) and Johnson et al. (2010) found, using data on Scottish children born in 1921, father's social class to be more important for educational attainment than child's IQ at age 11, a result that may reflect the difficulty for bright working-class children to attain more than compulsory education in the 1930s, a difficulty that could have been present also in Sweden at that time. The restrictions on advancing in the educational systems of both Scotland and Sweden seem recently to have become less severe, possibly due to the expansion of the educational system (see however Jonsson and Erikson 2007).

To use only one of the manifold possible indicators of social position to establish the association between social background and children's educational attainment may result in underestimating the overall association between background and attainment and overestimating the importance of the chosen factor. This is so because all sensible indicators will be correlated with one another, while the mechanisms that relate social background to educational attainment can to some extent be assumed to differ between aspects or dimensions. As a consequence, different parental characteristics may independently affect children's educational attainment. Thus, while for example parents’ education may influence children's educational aspirations, parents’ social class and income may have a more specific effect on learning facilities and the opportunities to choose a good school. To try to get as wide as possible a spectrum of children's social background, I will look at the simultaneous association between parental education, social class, social status and earnings, and children's education.

Children's educational attainment is assumed to depend on several background factors, which to some extent are transmitted via their cognitive ability. This assumption leads to two general questions:

It is not the intention of this paper to present an analysis of causal mechanisms, for the obvious reason that, with the exception of cognitive ability, I do not have measures of intervening variables that may identify these mechanisms. My hope is to provide a sound basis for developing hypotheses of causal processes by means of an accurate and reliable description of the connection between social background and educational attainment, which is a prerequisite for developing causal hypotheses.

The analyses were conducted on data from four national cohorts of pupils born during 1948, 1953, 1967 and 1972 collected by the Department of Education, University of Gothenburg (Härnqvist 2000). Ten per cent random samples were drawn of pupils in grade 6, when nearly all of them were 13 years of age. With the exception of the measure of cognitive ability, all data for this paper are based on register information from censuses and official registers, providing information of high accuracy for practically all pupils. Information about parental education and occupation was obtained from the 1960–1990 censuses and from the education register starting in 1985, while parents’ earnings were taken from the tax registers of 1968–1989. Information about the highest level of education the children had attained by the age of (in most cases) 32–40 was obtained from the education register.

Parents’ education is coded in relative terms, that is, as the proportion of parents in each cohort with less education than the index person's parents, a coding that facilitates the possibilities to compare the role of parental education for different cohorts (Bukodi and Goldthorpe 2013).¹ The educational measure thus varies from zero to just less than one. If parents’ educational levels differ, the higher level is used.

There are seven classes of parental EGP – a commonly used class schema (I, higher professional and managerial occupations, coded 1; II, lower professional and managerial occupations, coded 0.75; III, non-manual and service occupations, IVab, small employers and own account workers and IVc, farmers, all three coded 0.5; VI, lower supervisory and skilled manual workers, coded 0.25; VII, unskilled manual workers, coded 0) (Erikson et al. 1979). Furthermore, farmers (IVc) and other self-employed parents (IVab) are included in the models as two separate dichotomies in order to account for the specific conditions in these two classes of origin. The dominance principle (Erikson 1984) is used to determine parental class, if it differs between the parents. Parental status is based on the similarity between the occupations of married and cohabiting partners in the Census of 1990, as reflected in a CAMSIS scale (Prandy and Lambert 2003).² Status, as used here, is close to Weber's (1972/1921) concept of Ständische Lage, based on perceptions of equality, superiority and inferiority in social standing (cf Chan and Goldthorpe 2004). Parents’ log earnings are based on up to five-year averages. Parental status and earnings are, like the other independent factors, set to vary between zero and one. All background factors refer as closely as possible to the time when the pupils were 12–13 years of age. Children's education refers to the highest level attained on an eight-grade scale (Elementary, coded 1; Lower secondary, 2; Vocational upper secondary, 3; Academic upper secondary 4; Post-secondary, 5; University, 3 years 6; University, 4+ years, 7; Post-graduate, 8). The distributions of parental characteristics are presented in Table 1 and those of children's education in Table 2.

The measure of cognitive ability, based on identical tests administered to the four cohorts alike, is the sum of 120 items from three separate tests of verbal (opposites), spatial (metal folding) and reasoning factors (number series) (Emanuelsson et al.1993). Opposites refer to finding the opposite of a given word from among four alternatives. The test consisted of 40 items to be completed in 10 minutes. Metal Folding refers to finding the three-dimensional object among four alternatives that can be made from a flat piece of metal with bending lines marked on the drawing. Forty items were to be completed in 15 minutes. Finally, pupils had to complete series of eight numbers of which two were missing. Forty items should be completed in 18 minutes.³ The tests were taken by the children before they were separated into different secondary school tracks, which implies that the effect of schooling on cognitive ability should only to a very minor degree differ by social origin. To take any possible Flynn effect (Flynn 1987) into account, cognitive ability is set to vary between 0 and 1 for each cohort, respectively.⁴

Cognitive ability is presumably dependent on the interaction of environmental and hereditary factors – from the conception and onwards. Thus, cognitive ability at age 13 is partly dependent on the social background and the children‘s experiences up to that age. However, whatever the basis of cognitive ability, the question here is to what degree it, as it appears by the age of 13, mediates the association between social background and children's education.

All analyses are based on children's education regressed on background factors. By using Ordinary Least Squares Regression (OLS) it is possible to estimate the role of cognitive ability in mediating the total effect of social background, measured by four different factors. As the scale for parents’ class is an arbitrary classification of ordinal data, I checked whether the linearity assumption is severely inappropriate. No serious distortion of the association appears to result from regarding children's education as linearly dependent on parental class measured with the arbitrary scale.⁵ In the analyses, as mentioned above, all independent factors were set to vary between a minimum of 0 and a maximum of 1.

$βjryxj$⁠, the standardised regression coefficient for the independent factor $xj$ times the zero order correlation between $xj$ and the dependent variable y is equivalent to the relative importance of $xj$ independently of the other factors. That is, $βjryxj$ shows the contribution of $xj$ to the coefficient of determination (R²) and accordingly $Σβjryxj$⁠, the sum for all the independent factors, is equal to R². As R² can be interpreted as the proportion of variance in y accounted for by the linear combination of the independent factors, $R2βjryxj$ will indicate the proportion of variance in y accounted for by $xj$ independently of the other factors.⁶

All estimates of a factor's relative importance for the variance in children's education, that is, $R2βjryxj$⁠, relate to the same quantity – the variance in children's educational attainment – and may therefore, given that each model is based on the same data, be compared within models as well as between models including and not including cognitive ability. Accordingly, the effects of parental factors in models without cognitive ability can be compared with the corresponding effects in models which include cognitive ability. Thus, we can estimate how much of the association between the background factors and education is transmitted via cognitive ability and how much is ‘direct’, by dividing the sum of $R2βjryxj$⁠, over the background factors in a model with cognitive ability with the corresponding sum from a model without it.

The first question raised above concerns whether anything is gained by looking at several aspects of social background. In Table 3, R² for the association between each separate background factor and children's education is reported for all cohorts and so is R² from regression models where all background factors are included in the model as well as R² from a model with only cognitive ability as regressor. The corresponding results from a common model without control for cohort are likewise reported.

Of the four background factors, parental class and parental education display the highest overall correlations with children's education for both men and women, although parents’ social status seems to be of about similar importance as parental education in the two earlier cohorts. Parents’ log earnings show the lowest association throughout. The squared correlations vary from 0.04 to 0.16 among men and from 0.02 to 0.13 for women, thus indicating a lower association between background and education among women than among men.

Models that include all four factors account for between 14 and 15% of the variance among women and between 16 and 20 % among men. These results suggest that the four background factors, apart from being correlated with an implied latent stratification dimension, have separate additive effects. To include all four factors in the same model will make it possible to judge the relative importance of each factor, respectively.

Cognitive ability at age 13 accounts for around one fourth of the variance in education, and thus more than social background. The importance of cognitive ability seems to have decreased slightly, suggesting that the selection to higher levels of education in terms of ability was more rigorous in the early cohorts.

Parents’ education and their log earnings seem to have become more important for children's education between the cohort of 1948 and that of 1972. In spite of this possible change in importance of the background factors, the pattern over the years appears similar enough to motivate a simplification of the analysis by using models common for the four cohorts, that is, without control for year of birth.⁷

In Table 4 the R-squared for the common values in Table 3 are reported anew together with Pratt measures from models including one background factor and cognitive ability. As indicated above, it is possible to ascertain how much of the association between the background factors and children's education that is accounted for by cognitive ability by comparing a factor‘s contribution to R² in a model with cognitive ability to the corresponding R² in a model without it. While the direct effect of cognitive ability is only slightly reduced when comparing a model with cognitive ability only with one which also includes a background factor – from 80% for status among men to 96% for earnings among women – the effect of each background factor consistently falls by about one third when cognitive ability is included in the model. A similar result has previously been reported for Britain (Bukodi et al.2014).

The results obtained from regressing children's educational attainment on all the parental characteristics together are reported for men and women in all four cohorts in the upper panels of Table 5, while the results from models which include cognitive ability are reported in the lower panels. Parental education, class, status and earnings all have independent positive effects on children's educational attainment for both men and women (cf. Bukodi and Goldthorpe 2013; Bukodi et al.2014). Parental self-employment has no or possibly a weak negative effect for men and no significant effect for women. The effects are similarly weak or non-significant for farming parents, being on the whole negative for men and positive for women.⁸ Overall the coefficients are similar from one cohort to another, even if there is a certain increase in the effect of parental education and a decrease in the importance of class for women.⁹ Cognitive ability at age 13 obviously has a clear effect on educational attainment also given parental background, while the background factors remain important although reduced to around two-thirds of their magnitude in the models with cognitive ability. In Table 6 measures of relative importance and direct effects are reported based on common models without control for cohort.

In the first two columns of Table 6, Pratt measures from model 1, not including cognitive ability, are shown.¹⁰ Parental education and parental class both account for 5–7% of the variation in children's educational attainment and are clearly more important than parental status and earnings, which account for 1–4%. The order of the relative importance of the factors remains when cognitive ability is included in model 2, as is shown in the next two columns of Table 6. The relative importance of cognitive ability amounts to close to two-thirds of the variance accounted for by model 2, with parental factors accordingly accounting for slightly more than one third.

Together, the background factors account for 16–19% of the variance in educational attainment, as indicated by the R²s of Model 1. When cognitive ability is included in model 2, around or slightly less than one third of the variation in children's education is accounted for, as shown by the results for Model 2.¹¹ Cognitive ability at age 13 accounts for around 20% of this variation, while the background factors account for some 11–13% of the variance independently of cognitive ability, leaving around two-thirds of the variation in children's education unaccounted for.

The last two columns in Table 6 present the proportion of the association between each background factor and educational attainment that is ‘direct’. Overall, one third to one fourth of the effects of the background factors on educational attainment are transmitted via children's cognitive ability, while consequently more than two-thirds are ‘direct’, that is, transmitted via other mechanisms. Cognitive ability is least important for the association between parental earnings and children's education. The overall ‘direct’ effect of parental background for men and women is estimated to around 70% of the total effect.

Social background accounts for about one sixth of the variance in education and cognitive ability at age 13 for around one fourth, while they together account for around one third. Of this ‘explained’ third, cognitive ability accounts for slightly less than two-thirds and social background for around one third. These approximate results hold for both women and men. Close to 30% of the effect of social background is channelled through cognitive ability. Thus over 70% of the effect of the social background on education is transmitted via other ways than cognitive ability at the age of 13. Further conditions connected to the social positions of parents like non-cognitive abilities, preferences and perhaps the geographical location of the parental home are thus important for children's educational attainment.

All the four background factors – parental education, class, status and earnings – independently influence children's educational attainment both ‘directly’ and indirectly via cognitive ability. The direct effect of each background factor amounts to around two-thirds of the total effect of the factor. Parental education and class appear to be relatively more important than the other two background factors: together they account for 11–13% of the variance in attained education for both men and women, while status and earnings together account for 5–6%. The results show that Lazarsfeld's suggestion is too restrictive, a latent variable cannot account for all effects of the social background.

The basic association between the background factors and educational attainment changes very little over time. This is in line with earlier results from Sweden which suggest that social selection in education decreased among cohorts born before 1950, but remained rather stable among later cohorts (Erikson and Jonsson 1996).

These analyses show that cognitive ability at age 13 is important for educational attainment, but cognitive ability is far from being the basic mechanism. A similar result was found in analyses of primary and secondary effects, where secondary – that is, direct – effects based on cognitive ability at the age of 13 account for around two-thirds of the social selection to academic upper secondary education (Erikson and Rudolphi, 2010). These findings clearly contrast with the expectations of Burt and also differ from Nettle (2003), who reported that most of the association between parental social class and attained social class was mediated through general ability. The contrast is further accentuated by the fact that cognitive ability is with all probability more important for the attainment of education than for that of occupation. On the other hand, the results are in line with those of Deary et al. (2005) and Mood et al. (2012), who found that cognitive ability has an important but nevertheless limited effect on the association between social background and attained education.

What do the results presented in this study suggest about the possible mechanisms, apart from cognitive ability, which link background and education?

There may be an effect of stratification in general, as indicated by each specific factor influencing children's educational attainment. Marmot (2004) suggests that the status order in itself, by placing people on different levels of social standing and accordingly on different levels of subordination and economic and social conditions, affects their health and life expectancy. A corresponding effect could be hypothesised for educational attainment and could operate for any indicator of social position. However, specific mechanisms related to the separate factors seem also to be important for the direct effect of parental background.

Education is presumably more highly valued in highly educated families, which should lead both parents and the children themselves to aspire for the child to stay in school longer. Breen and Goldthorpe (1997) assume that families and children find it more important to avoid downward mobility than to progress upwards – an assumption which is in accordance with prospect theory (Kahneman and Tversky 1979). Furthermore, parents with a higher level of education will on average be better able to help their children with their school work, a factor that, at least in Sweden, may actually have become more important in recent years, since more emphasis has been placed on the role and support of parents during their children's early school years. Knowledge of the intricacies of the educational system may also make more educated parents better able to give their children good advice at decision points during the school career. The children of more highly educated parents can be expected to acquire a more elaborate vocabulary (Becker 2011) and be more likely to find reading a normal and common activity at home. Their parents may also be more inclined to read to them during the infant years, all of which could result in more and improved reading among the children and to an improved language development (Sénéchal et al.1996). The variation in the verbal interaction between parents and children may influence children's cognitive development and thus, rather than only being a direct effect, may also influence the pathway from parental education to children's education via cognitive ability.

Assumptions about the amount and kind of education needed to get what is regarded as a good job may differ by social class. That is, while a family with professional occupations may regard higher education as a necessity, a working-class family may regard vocational training as the best way forward (cf. Breen and Goldthorpe 1997). The possible learning of specific skills at home could have a similar effect. If children of self-employed parents were expected to take over the firm or the farm, on-the-job training may, in the period covered by the data, have been regarded as more important than further schooling. Furthermore, the better economic resources of salaried parents should improve the educational prospects of their children in a number of ways. The higher degree of economic security and the more foreseeable economic future of salaried families may be of particular importance (cf Erikson 1996). Greater economic resources will presumably improve children's prospects of a successful educational career by making it easier to cover the costs of education, by providing children with more space and better facilities for their school work and, not least, by enabling the family to choose to live in a neighbourhood with high-quality schools.

Socio-cultural resources and the culture at home will presumably differ by social status. In other words, the cultural values in families of higher status will be closer to those typically promoted in the academic tracks at school. The composition of the parental network may be important for parents’ expectations of and attitudes towards their children's educational careers (Roth and Salikutluk 2012).

Different aspects of the social background as well as cognitive ability affect children's educational attainment. Cognitive ability at age 13 accounts for more of the variance in education than social background, but all the four background factors – parents’ education, social class, social status and earnings – have effects independently of each other and of cognitive ability. Most of the effects of these factors are transmitted via other mechanisms than cognitive ability. Cognitive ability is thus an important link between social background and educational attainment, while far from being the major mechanism behind the association. This presumably applies even more to social mobility, given the assumption that cognitive ability is more important for educational attainment than for the achievement of social class.

Parental education and social class are most important of the four background factors, together accounting for more than 10% of the variance in children's education. Among mechanisms accounting for these ‘direct’ effects, parents’ education and status may be important for how education is valued in the parental home, while the variation in economic resources may stand for most of the effects of parental class and earnings.

Social background, conceived of as composed of several related but distinctive factors, accounts for more of the variation in educational attainment than any of the factors separately. However, regardless of the size of the relative effects of these factors, cognitive ability at age 13 accounts for about one third to one fourth of the association between each background factor and educational attainment.

In conclusion, returning to the questions raised at the beginning of the paper, it would appear that:

Martin Hällsten, Michael Tåhlin, Nanny Wermuth and a reviewer have provided help that clearly improved the quality of the paper.

No potential conflict of interest was reported by the author.

Robert Erikson is professor of sociology at the Swedish Institute for Social Research, Stockholm University, presently studying the role of education in the intergenerational transmittance of advantage.