Longitudinal matching of recent current population surveys: Methods,non-matches and mismatches

Abstract

Longitudinal matching of Current Population Survey data is of considerable interest to empirical researchers. This paper develops a general method that uses the conditional probabilities (of being correct) to accept or reject a crude match. The assumptions and procedures used to estimate those probabilities with real data are described in detail. The paper performs the matching of adjacent recent March CPS Supplements, and finds a sharp increase in non-match rates after the 2002 State Children Health Insurance Program (SCHIP) sample expansion. The widely used matching algorithm proposed by Madrian and Lefgren continues to work fine for recent CPS, as it keeps the mismatch rates low in all years.