A variety of randomized response (RR) procedures for privacy and condentiality protection have been proposed, studied and compared in the literature. We describe statistically relevant attributes of RR mechanisms and use those to duly organize and unify existing estimation theories for diverse RR methods. Any RR procedure can be characterized by its transition probability matrix P, as it determines all statistical properties of the procedure. In RR surveys, P is xed, but in post-randomization for confidentiality protection, P may depend on the original data. This affects statistical inferences significantly. We also review some optimality results in the comparison of RR surveys of a binary variable, based on both privacy protection and statistical eciency.