Treatment effect without controls—different sample restrictions. Panel (a): academic high school completion. Panel (b): university enrolment or graduation Notes: This figure shows a sensitivity analysis where we sequentially drop imbalanced classes and then estimate treatment effects. This figure shows the relation of shrinking the estimation sample by excluding classes with highly imbalanced shares of fathers who completed academic high school and estimated treatment effects on the outcomes academic high school completion at age 20 in Panel (a) and university enrolment or graduation at age 24 in Panel (b). The x-axis shows the number of classes that are dropped. Because of the trimming mechanism, we always drop two classes simultaneously. These are the current classes with the lowest share of fathers who completed academic high school in the treatment group and the current class with the highest share of fathers who completed academic high school in the control group. We only consider classes with at least six students. The y-axis shows the point estimate when regressing the treatment indicator PATHS on the outcomes academic high school completion and university enrolment or graduation. The blue diamond shows the specification in which the imbalance of fathers with at least a Gymnasium degree between the treatment and control groups is closest to zero. All models include strata fixed effects for the level of randomization. Robust standard errors are clustered at the school level.