Communication with partially verifiable information : an experiment

We use laboratory experiments to study communication games with partially verifiable information. In these games, based on Glazer and Rubinstein (2004, 2006), an informed sender sends a two-dimensional message to a receiver, but only one dimension of the message can be verified. We compare a treatment where the receiver chooses which dimension to verify with one where the sender has this verification control. We find significant differences in outcomes across treatments. Specifically, receivers are more likely to observe senders' best evidence when senders have verification control. However, receivers' payoffs do not differ significantly across treatments, suggesting they are not hurt by delegating verification control. We also show that in both treatments the receiver's best reply to senders' observed behavior is close to the optimal strategy identified by Glazer and Rubinstein.