Publication Date





To compare different schemes for preserving privacy, it is important to be able to gauge loss of privacy. Since loss of privacy means that we gain new information about a person, it seems natural to measure the loss of privacy by the amount of information that we gained. However, this seemingly natural definition is not perfect: when we originally know that a person's salary is between $10,000 and $20,000 and later learn that the salary is between $10,000 and $15,000, we gained exactly as much information (one bit) as when we learn that the salary is an even number -- however, intuitively, in the first case, we have a substantial privacy loss while in the second case, the privacy loss is minimal. In this paper, we propose a new definition of privacy loss that is in better agreement with our intuition. This new definition is based on estimating worst-case financial losses caused by the loss of privacy.