Regression-based Statistical Matching: Recent Developments

• Activity Number: 388
• Type: Contributed
• Date/Time: Wednesday, August 11, 2004 : 2:00 PM to 3:50 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Chris Moriarity*+ and Fritz J. Scheuren
• Companies: U.S. General Accounting Office and NORC, University of Chicago
• Keywords: data fusion

Abstract

We have described a method in several articles (2001, 2003) for statistically matching two samples. One sample is assumed to contain (X,Z) and the other is assumed to contain (X,Y), both drawn from a common nonsingular normal (X,Y,Z) distribution. Following Kadane (1978) and Rubin (1986), we employ regression in our approach. We assess the uncertainty introduced during the match that is due to the unobserved (Y,Z) relationship by repetition over a range of (Y,Z) values that are consistent with the observed data. In the final step of our algorithm, we replace predicted values with observed data by a match of the two samples to obtain final files consisting only of observed data, consistent with traditional statistical matching procedures. Prior to matching, we add random residuals to our regression-based estimates, an essential step in our method. Our approach for estimation of the amount of residual to add, using subtraction and estimates from both files, can be negative. Rassler (2002) suggests a different approach for residual estimation, which is always non-negative. We compare the two methods and discuss other recent developments.