Working Paper Series, UCD Centre for Economic Research 10/05
Objective: To estimate the causal effect of breastfeeding on children's cognitive skills as measured at ages 3, 5, 7 and 11. Design: An instrumental variable (IV) strategy which provides a correction method for dealing with selection bias. Standard linear regression models are compared to two-stage least squares models to test for the presence of endogeneity. The consistency of the results across multiple sources is also tested using data from two prospective longitudinal studies collected 40-years apart. Setting: The 1958 National Child Development Study (NCDS) and the 2000 UK Millennium Cohort Study (MCS). Participants: Data on 11,792 (age 3) and 9117 (age 5) children in MCS and 4923 (age 7 and 11) children in NCDS. Main outcome measures: Cognitive ability is measured by the Bracken School Readiness Assessment (age 3); Foundation Stage Profile (age 5); and tests of general ability including mathematics, comprehension, verbal and non-verbal skills (ages 7 and 11). Results: The duration of breastfeeding has a small, but significant, effect on children's cognitive skills in the linear regression models at ages 3, 5, 7 and 11, but no effect in the IV models. However, in all cases, the hypothesis that breastfeeding is endogenous is rejected, indicating that the results of the linear regressions are valid. Conclusion: The relationship between breastfeeding and cognitive ability is not driven by selection bias once a rich set of confounders are included. IV methods can therefore be used to test for the presence of selection bias and are a useful alternative for identifying causal relationships when randomised control trials are not feasible. Showing that the size of the effect is similar for two cohorts born over 40 years apart, and using different measures of ability, are further indications that the relationship between breastfeeding and cognitive ability is not a statistical artefact.