The intergenerational effect of parental education on child health: evidence from the UK

Abstract

While many earlier studies have shown a positive correlation between parents' education and children health, little attempt has been made to address the possibility that unobserved characteristics underlie this intergenerational relationship. This paper explores the effect of additional schooling induced through compulsory schooling laws in Great Britain and Northern Ireland on child health as measured by long-term illness. Despite statistically significant ordinary least squares estimates, the instrumental variable estimates reveal little if any causal relationship between parental education and long-term illness in children. This is true for both two-parent and single-parent families.

Keywords:

• education
• health
• endogeneity bias

JEL Classification:

• I21

Notes

1 See, for a comprehensive review of this literature, Grossman (2006).

2 Silles (2009) only used data for Great Britain. The analysis here is extended to Northern Ireland particularly in an attempt to increase sample size and variation in the educational instrument.

3 For example, in 1971 there were an estimated 1 million dependent children in 0.57 million single-parent families. These numbers rose to 2.8 million children in 1.6 million one-parent families by 1996 and continued to rise to 3.2 million children in 1.9 million single-parent families by 2012 in the UK.

4 Silles (2011), using the GHS/CHS data-set and compulsory school reforms as instruments for education, found that increases in schooling reduce the probability of teenage pregnancy. However, these data do not contain completed life cycle fertility histories for young women or retrospective fertility information for older women whose children may have left the household. This information would be necessary to permit a more complete investigation of the hypothesis that fewer births lead to better child health.

5 Scotland has a different system of schooling from England and Wales and therefore is dropped from the analysis.

6 Marginal effects from probit binary dependent variable models produce similar results.