3.8  Children’s education

Studies throughout the world find that better-educated people tend to have children who experience better outcomes: their children are healthier, for example, do better at school and commit fewer crimes. Young people in the Christchurch HDS cohort whose mothers had low levels of education were at greater risk of many poor outcomes in childhood and adolescence.[15] It may be, therefore, that an increased education benefits not only the individual who receives it but also their children. This section looks at evidence for one particular causal relationship, that between parents’ education and their children’s education.[16]

Haveman and Wolfe (1995) review studies on the family and neighbourhood determinants of children’s years of education and of high school graduation. They find that in virtually all of these studies, parents’ education is statistically significant and quantitatively important, no matter how it is defined. This is the case despite controlling for other parental, family and neighbourhood characteristics such as income, occupation, family structure, and maternal age. Mothers’ education was found to be more closely related to the attainments of the children in these studies than fathers’ education. There are plausible explanations for these relationships. More educated parents, and particularly mothers, might be better at encouraging their children to study, setting high expectations, being role models of possible careers, reading to their young children and helping them with their homework.[17] Better education of children might therefore begin a cycle, where succeeding generations also do better at school.

These explanations ignore the effect of genetic inheritance, however. It might be that more naturally able parents do better at school and that this natural ability is in turn genetically transmitted to their children, who also do better at school.[18] Well-educated mothers might also marry more able men, and as a result of this assortative mating might improve the genetic endowments of their children. Adoption studies, twin studies, and studies of increases in education by young mothers all attempt to control for these genetic factors.

Adoption studies

Adoption studies take advantage of the fact that adopted children share only their parents’ environment, and not their parents’ genes. Any relationship between the education of adopted children and their adoptive parents is therefore driven by the influence that parents have on their children’s environment, and not by parents passing on their natural ability. Sacerdote (2000) and Plug and Vivjerberg (2001) both find that parents’ education has a sizeable and significant effect on their adopted child’s education. An additional year of mother’s education, for example, is associated with an extra 0.22 years of schooling for the child (0.26 years in Plug and Vivjerberg). Children adopted by a mother with a college degree have a 40% higher chance of graduating from college themselves (19% in Plug and Vivjerberg). Both studies control for basic demographic variables.

Sacerdote also looks at the effect of parents’ education on their adopted child’s test scores, as well as on their quantity of schooling. He finds that the effect on test scores is relatively small, suggesting that family environment is able to influence educational attainment and college attendance to a greater degree than test scores. This result suggests that test-taking ability relies more heavily on genetic endowments whereas college attendance relies more heavily on parental expectations. Plug and Vivjerberg examine the effects of assortative mating, by controlling for the effect of mother’s and father’s education together. This did not influence the relationship between the education of fathers and their adopted children, but it strongly reduced the relationship between education of mothers and their adopted children to the point of statistical insignificance. This result suggests that a mother’s education improves her child’s schooling indirectly, through her choice of marriage partner. Plug and Vivjerberg also include family income in their models, but this has little effect on the coefficients, suggesting that education has a direct effect rather than acting through increased income.

Adoption studies tend to be hampered by small sample sizes, since relatively few children are adopted. Sacerdote uses a sample of only 170 children from the NLSY; Plug and Vivjerberg use a sample of 610, although 40% were still at school and had not completed their education. Sacerdote also admits the possibility that adopted children may not have been randomly assigned to parents and that high-ability parents may have been able to select, or may have been matched to, children from high-ability birth mothers. Furthermore, neither of these two studies controls for characteristics of parents which might be associated with their own educational performance and also with their parenting ability – characteristics such as patience, communication, diligence, and intelligence. For this reason, the effect of parents’ education on their adopted children might well be overestimated in the adoption studies discussed above.

Twin studies

Twin studies take advantage of the fact that identical twins raised together are extremely similar: they share 100% of their genes, and grow up in the same family.[19] Behrman and Rosenzweig (2002) look at the difference in the years of education completed by pairs of identical twins, and then relate this to the difference in years of education completed by their eldest children (who are therefore cousins). Identical twins are assumed to have the same natural ability and also the same innate childrearing ability.

Surprisingly, Behrman and Rosenzweig show that the child of the more-educated sister in each pair of female twins tended to complete less education. A one-year increase in a mother’s education was associated with a decrease in her child’s education of about a quarter of a year, although this was only significant at the 10% level. On the other hand, the child of the more-educated brother in each pair of male twins tended to complete more education. A one-year increase in a father’s education was associated with a statistically significant increase in the child’s education of about a third of a year. For both female and male twins, adding their husband’s or wife’s education and ability into the model, and therefore controlling for assortative mating, had little effect on the observed relationships.

These findings suggest that increasing men’s education would raise the level of education of the next generation by a small amount, while raising the level of education of women would not, and might even lower it. Behrman and Rosenzweig consider that these findings might be explained by reference to women’s time in the home: women with more education tend to spend less time at home with their children, and this negatively affects their children’s learning. As with adoption studies, however, sample size is an issue which plagues twin studies. Behrman and Rosenzweig claim to have used a sample of 212 female and 122 male twin-pairs but Antonovics and Goldberger (2003) allege that relatively few twin-pairs in the study had children who had actually completed their education.

Studies of increases in mothers’ education

Two studies take advantage of the fact that young mothers sometimes return to formal education after the birth of a child or between the births of their first and subsequent children. Both studies use an extension of the NLSY which surveys the children born to women in that cohort.

Kaestner and Corman (1995) associate young children’s improvements on tests of reading and mathematics, two years apart, with increases in their mother’s formal education over this period. They control for other changes in the children’s lives over this period which might have affected their test results, such as their health status, mother’s labour force participation, family structure and family income. Factors which did not change over this time, such as the children’s ability or their mother’s innate parenting ability, were common to both points in time and therefore implicitly controlled for in the model. Kaestner and Corman find no effect of increased maternal education on children’s achievement scores.

Rosenzweig and Wolpin (1994) look at differences in test scores between earlier-born and later-born pairs of siblings, relating these differences to increases in their mother’s formal education over the intervening period. A quarter of the sample had continued their education after the birth of their first child. Rosenzweig and Wolpin control for a number of factors which might vary between the earlier- and later-born children, and which might affect cognitive development, such as the use of prenatal care, birth-weight, mother’s smoking and drinking, home teaching, television watching and frequency of reading to the child. Family and parental factors were common to both children and therefore implicitly controlled for in the model (although the children, not being identical twins, might have differed in other unobservable ways). Rosenzweig and Wolpin find that an additional year of maternal education has a modestly positive and marginally significant effect on their children’s achievement in reading and mathematics tests, although not on a measure of verbal IQ. Each additional year of education obtained by the mother prior to the birth of a child increases achievement test scores by 2.4%.

Both of these studies have reasonable sample sizes to work with, but both consist of relatively young women and their children. The oldest respondent in Kaestner and Corman’s study at the time of the second assessment was only 34 years old, with a child aged between 7 and 9. Rosenzweig and Wolpin’s sample was made up of women who had had at least two children by age 25 (for the maths and reading sample) or by age 27 (for the vocal IQ test). Half of these mothers had first given birth between the ages of 17 and 19, which explains their high rates of school continuation. The samples are therefore not representative of all families with children and this limits the generalisability of the results.

Instrumental variables studies

Two recent NBER papers use changes in compulsory schooling laws as an instrument for parents’ education, in an attempt to determine whether an increase in parents’ attainment affects their children’s attainment. Historical changes in compulsory schooling laws are used because they affect the length of time that parents are at school without affecting their innate abilities. Black, Devereux and Salvanes (2003) look at the effect of an increase in the duration of compulsory schooling in Norway from 7 to 9 years. This reform was introduced in different municipalities at different times between 1960 and 1972. Black et al use these differently-timed law changes as an instrument for parental education, and find little evidence of a causal relationship between the length of fathers’ education and the length of their children’s education. They do, however, find a small, but statistically significant, relationship between a mother’s education and her son’s (but not her daughter’s) education. These findings suggest that the high correlations between parents’ and children’s education are due primarily to family characteristics and inherited ability.

Oreopoulos, Page and Stevens (2003) come to a different conclusion. Their study uses changes in compulsory schooling laws across different states of the United States as an instrument for parental education, and they apply this instrument to data from the 1960, 1970 and 1980 Censuses. In contrast to Black et al, Oreopoulos et al find that a one-year increase in the education of either parent reduces the probability that a child repeats a grade at school by between two and seven percentage points. Among 15 to 16 year olds living at home, Oreopoulos et al also estimate that parental compulsory schooling significantly lowers the probability of dropping out of high school. These findings suggest that an increase in parents’ education may cause an increase in their children’s educational attainment.

Summary

The evidence on the relationship between parents’ education and their children’s education is conflicting. Most of the studies discussed above find some evidence of a causal relationship, but differ as to whether this applies to both parents (e.g. Oreopoulos et al 2003), to fathers only (e.g. Behrman and Rosenzweig 2002), or to mothers only (e.g. Black et al 2003). In addition, the evidence from adoption studies, twin studies, and studies of increases in education by young mothers is difficult to interpret because of methodological complexity and problems such as the small sample sizes in most of the studies.