Figure 1.

Mean two‐stage least squares/inverse‐variance weighted estimates plotted against sample overlap for different values of instrument strength ($\alpha =0.4$, circle; $\alpha =0.6$, triangle; $\alpha =0.8$, plus) and different values of the confounder effect on the outcome (${\beta }_U=0.6$, black solid line; ${\beta }_U=1$, mid‐gray dashed line; ${\beta }_U=2$, light‐gray dotted line). Left panel: positive causal effect (${\beta }_X=0.2$); right panel: null causal effect (${\beta }_X=0$)