Roth studies algorithm design in settings in which either the data belongs to other self-interested parties, or the computation is to be performed by other self-interested parties. This requires studying the algorithmic foundations of data privacy and game theory. In particular, he is interested in what kinds of computations can be performed while satisfying strong information-theoretic privacy constraints, and in how selfish agents can be incentivized by these strong privacy protections to allow their data to be used.