This dissertation studies several applications of the recently developed information design framework. The first chapter introduces an opinion leader into a binary-state binary-action Bayesian persuasion model, who the sender must convince before his target audience is made aware of his experiment. I show that the presence of the opinion leader makes the sender's experiment weakly more informative. I also consider two forms of collusion, and show that while collusion between the sender and opinion leader may help or harm audience welfare, under collusion the audience weakly prefers the opinion leader to be present. The second chapter asks which Bayesian persuasion outcomes can be attained by a researcher who can only control the number of subjects to enroll in a trial. I show that under sequential sampling, as subject outcomes become arbitrarily uninformative, the researcher can induce any Bayesian persuasion outcome. However, under pre-registration, I show that the researcher is bounded away from his first-best Bayesian persuasion outcome. The third chapter concerns a refinement of Bayes correlated equilibrium in which each agent must not only be willing to follow the mediator's recommendation once he hears it, but must also be willing to hear the recommendation in the first place. I show for a fixed game that the set of ``hear-no-evil equilibria'' is shrinking in the level of players' baseline information if and only if information always benefits the agent who hears it. I also show that Pareto-optimal outcomes need not satisfy the hear-no-evil condition. For a class of 2x2x2 coordination games I also show that every symmetric BCE satisfies the hear-no-evil condition.