Abstract
This paper analyzes identification and statistical inference issues related to matched March Current Population Survey (CPS) data sets. Existing studies using matched March CPS typically ignore the non-match and mismatch problems, which is equivalent to imposing strong but not credible assumptions for point identification. I instead take a partial identification approach, starting from identification regions based on empirical evidence alone, followed by explorations of the identification powers of more credible assumptions, some of which are directly motivated by the understanding of the matching process. The paper uses U.S. poverty dynamics as an example to illustrate the method proposed.