Protocols for information-hiding often use randomized primitives to obfuscate the link between the observables and the information to be protected. The degree of protection provided by a protocol can be expressed in terms of the probability of error associated to the inference of the secret information. We consider a probabilistic process calculus approach to the specification of such protocols, and we study how the operators affect the probability of error. In particular, we characterize constructs that have the property of not decreasing the degree of protection, and that can therefore be considered safe in the modular construction of protocols. As a case study, we apply these techniques to the Dining Cryptographers, and we are able to derive a generalization of Chaum's strong anonymity result.