Fig. 6

Sensitivity analysis of the estimated effect of selective schooling on the health outcomes from hospital admission and death records. Analyses are (1) using higher weights for those closer to the cut-point (triangular); (2) using equal weights; (3) including confounders of sex, IQ at age 9, and father’s occupation; (4) including the 3048 people in the base sample for whom secondary school was not known using inverse-probability weighting; (5) excluding those whose probability of attending selective secondary was not 0 or 100% (donut analysis); and (6) including the entire population equally weighted. Each of these variations was evaluated for first-, second-, or third-order polynomials of the 11+ exam score. Each sensitivity analysis effect estimate is represented as a point with associated 95% confidence intervals