This paper proposes a new method called ‘cross-screening’ to increase the power of sensitivity analysis when multiple causal hypotheses need to be tested simultaneously.