Improving the Validity of Causal Inferences in Observational Studies