Sensitivity check—Difference of effects under different confounding scenarios relative to the baseline scenario (ATE).

Note: Each cells shows the difference in the effects resulting from an alternative confounding scenario to the baseline scenario. Scenarios 1–3 simulate the new confounding variable mimicking the relation of three important confounders to treatment and outcome, respectively (scenario 2 mimics the relation to family income > 5000 Euro; scenario 3 mimics the relation to mom’s education degree being university, scenario 4 mimics the relation to the local employment rate). Inference is based on 99 bootstrap replications and 19 draws of simulated binary confounder; quantile method, smoothed version, linear / logistic bias adjustment, symmetric p-values used. None of the differences is significant at the 10% level.