How and Why to Use Experimental Data to Evaluate Methods for Observational Causal Inference