Biased Recommendations

We develop and experimentally test a discrete choice model of an expert who recommends one of multiple actions to a decision maker who might take no action. Consistent with the recent theoretical literature on cheap talk recommendations, we find that recommendations are "persuasive" in that they reduce the chance that the decision maker takes no action, that recommendations for an action the expert benefits more from are "discounted", that lack of "transparency" about expert incentives undermines communication by both biased and unbiased experts, and that experts "pander" to the perceived preferences of the decision maker.