Francisco Silva;
Abstract: The literature initiated by Green and La¤ont (1986) studies principal/agent models with hard evidence. Evidence is modeled by assuming that the message set of the agent is type dependent. In this setup, Glazer and Rubinstein (2004, 2006) and Sher (2011) show that, when the agent’s utility function is type independent, there is no advantage for the principal in having commitment power. This paper shows that this way of modelling evidence implicitly assumes it to be perfectly accurate and that the result that commitment power has no value is not robust to making the evidence imperfect.