The commonly held view is that extremely poor health hurts educational achievement. This study examined the possibility of biases in standard estimation of effects and illustrated empirically, based on Ghanaian Living Standard of Measurement Study data, that there was not a significant effect of child health on child cognitive achievement. Consideration was given to the endogenous determination of child health. Child health was determined by anthropometry. Cognitive achievement test scores and preschool ability measured schooling success and child endowments respectively. Household and community characteristics and sibling data were used to measure family and community fixed effects. The conclusion, based on ordinary least squares (OLS) estimation and instrumental variable level estimates, was that child health did not impact on child cognitive achievement.^

The differences between the instrumental variable estimates and the within family and within community estimates suggested bias. Four basic conclusions were drawn. 1) Considerable bias occurred in prior studies, because there was a failure to account for estimation problems. 2) Inclusion of instrumental variables, assumed to be independent of the disturbance term in the cognitive achievement production function, and without controls for simultaneity, suggested a downward bias. 3) The bias was upward when estimates with sibling data were accounted for. Unobserved family and community effects can cause upward biases. 4) Coefficients, which are supposed to represent the impact of child health on schooling, may not do so.^

In the discussion of model specification, it was pointed out that upward bias can occur with heterogeneity in preferences regarding child quality, in unobserved predetermined family endowments that affect production of child quality, and in unobserved predetermined community endowments that affect child quality production. The unobserved predetermined child characteristics affect both child health and cognitive achievement in the same direction. There can be unobserved heterogeneity in access to capital markets. The 2 stage least squares procedure was found to overstate the impact of child health and cause greater distortions than OLS estimates.