The (Human) Sampler's Curses

We present a cheap talk model in which a receiver (R) sequentially consults multiple experts who are either unbiased or wish to maximize R's action, bias being unobservable. Consultation is costly and R cannot commit to future consultation behavior. We find that individual expert informativeness negatively relates to consultation extensiveness and expert trustworthiness due to biased experts' incentive to discourage further consultation by mimicking unbiased experts. We identify three (sampler's) curses: R may lose from an increase in the number or in the trustworthiness of experts as well as from a decrease in consultation costs.