False Positives and Transparency

I develop a theoretical model of costly information acquisition in order to evaluate transparency requirements in empirical research. A sender chooses an experiment characterized by multiple dimensions, while a receiver observes the experiment's outcome (though not necessarily all dimensions). I show that the receiver may prefer to keep dimensions hidden, even those contributing to bias, despite preferring more informative experiments. This can occur if the perception of bias is lessened when the sender compensates along a dimension that is observed. I elucidate how complementarity between dimensions underlies this result.