Incomplete-data problems arise naturally in many instances of statistical practice. One class of incomplete-data problems, which is relatively not well understood by statisticians, is that of merging micro-data files. Many Federal agencies use the methodology of file-merging to create comprehensive files from multiple but incomplete sources of data. The main objective of this endeavor is to perform statistical analyses on the synthetic data set generated by fileยญ merging. In general, these analyses cannot be performed by analyzing the incomplete data sets separately. The validity and the efficacy of the file-merging methodology can be assessed by means of statistical models underlying the mechanisms which may generate the incomplete files. However, a completely satisfactory and unified theory of file-merging has not yet been developed This monograph is only a minor attempt to fill this void for unifying known models. Here, we review the optimal properties of some known matching strategies and derive new results thereof. However, a great number of unsolved problems still need the attention of very many researchers. One main problem still to be resolved is the development of appropriate inference methodology from merged files if one insists on using file merging methodology. If this monograph succeeds in attracting just a few more mathematical statisticians to work on this class of problems, then we will feel that our efforts have been successful
CONTENT
1. Introduction and Summary -- 1.1 A Dichotomy of Matching Problems -- 1.2 A Paradigm -- 1.3 A General Set-up for Statistical Matching -- 1.4 The Statistical Matching Methodology -- 1.5 Constrained and Unconstrained Matching -- 1.6 Reliability of Synthetic Files -- 1.7 Summary -- 2. Merging Files of Data on Same Individuals -- 2.1 AGeneralModel -- 2.2 Notations -- 2.3 Model-based Matching Strategies -- 2.4 Repairing a Broken Random Sample -- 2.5 Reliability of Matching Strategies for Bivariate Data -- 2.6 An Optimality Property of the Matching Strategy ?* -- 2.7 Monotonicity of E(N(?*)) with respect to Dependence Parameters -- 2.8 Some Properties of Approximate Matching Strategies -- 2.9 Poisson Convergence of N(?*) -- 2.10 Matching variables with independent errors model -- 3. Merging Files Of Data On Similar Individuals -- 3.1 Kadaneโ{128}{153}s Matching Strategies for Multivariate Normal Models -- 3.2 Alternatives to Statistical Matching Under Conditional Independence -- 3.3 An Empirical Evaluation of Certain Matching Strategies -- Appendix A -- Appendix B -- References -- Author Index
