Thesis: This thesis consists of three independent essays, each of which focuses on a theoretical or empirical aspect related to expert-based knowledge. In the first chapter of this thesis, I study the transmission of scientific knowledge between an expert and a decision maker. A scientific model is formalised by a probability distribution over a set of possible scenarios. The expert is assumed to know the most likely model among a set of possible models and tries to communicate it to the decision maker. However, because these models are too complex, the expert cannot certify this information to the decision maker. I show that if there is a difference of interest between both parties, at equilibrium, the transmission of information is always partial. The expert will never be able to credibly communicate which model is the most likely. However, he will be able to designate a set of models containing it. The size of this set, and thus the degree of information that can be communicated, depends both on the difference of interest between the parties and on the consensus among scientific models. If the science is not sufficiently consensual, there is an asymmetry in the transmission of information. If the most likely model is among the most optimistic, the transmission of information depends solely on the difference in interest between the parties. But if it is among the most pessimistic, no information transmission is possible. In the second chapter of this thesis, my co-authors and I experimentally measure subjects’ beliefs about events with which they are more or less familiar. To do so, we propose a novel method for identifying subjects’ beliefs that relies on the use of objective probability intervals. For each event, our approach allows us to elicit mostly non-degenerate probability distribution sets. Moreover, we find that the more familiar the events, the smaller the elicited intervals. Thus, the more our subjects believe themselves to have expertise on a question, the more accurate their beliefs are. Our approach also allows us to estimate how these subjects act in correspondence with their beliefs. In doing so, we arrive at the first estimate of the mixture coefficient α in Hurwicz’s α-maxmin EU decision model, controlling for the subjects’ beliefs. In the third chapter of this thesis, I once again assume that scientific knowledge is too complex to be certified to non-experts. I study the consequences of this assumption in the case of climate change mitigation. I model the problem of greenhouse gas (GHG) over-emission as a game of contribution to a public bad. In this game, all contributors individually gain from emitting, because GHGs are correlated with the consumption of goods, but all contributors suffer from the total level of emissions because these emissions are responsible for climate damage. In equilibrium, the level of emissions is always too high, because each contributor does not take into account the negative externalities for which it is responsible. The contributors are not climate experts, and their knowledge of the damage they are exposing themselves to is based solely on an expert. The expert takes into account the externalities of the contributors, and would always want a lower emission level than the one obtained in equilibrium by the contributors. Thus, there is always a difference of interest between the expert and the non-experts. In this chapter I prove that no transmission of information can take place at equilibrium. This result shows that the word of the expert alone, without certification power, is not enough when it comes to communicating about climate risk.