Abstract
This paper proposes an algorithm to match Current Population Survey (CPS) respondents based on Bayes' rule. For each naive match, Bayes' rule is used to calculate conditional probability of being correct or not, given certain characteristics of the respondent. This conditional probability is then used to validate or invalidate the match. In an example to match CPS March supplements, Type I and Type II errors are derived for the proposed algorithm and the one by Madrian and Lefgren. The proposed algorithm is more efficient in that it tends to reject less correct matches. The paper also uncovers the recycling pattern of CPS household identifiers. Based on this recycling pattern, the proposed algorithm is also extended to the matching of CPS basic monthly data sets.