Replicating Experiments Using Surveys and Aggregate Data: The Case of Negative Advertising