Evaluating Nonexperimental Estimators for Multiple Treatments: Evidence from Experimental Data