Abstract:
|
For automatic data editing/imputation (E/I) with linear edits under the Fellegi-Holt methodology, the linear programming approach provides proper methods for error localization. However, in practice, the computer implementations of error localization procedures have not been very satisfactory in the performance. Various efforts have been made, including development of new methodology and faster new algorithms, to improve the situation.
In developing a linear edits error localization module for NASS' new processing system for its 2002 Census of Agriculture, we addressed the methodological issue of elimination and tried implementation in C++, using object-oriented design, to seek improvement of computational efficiency.
This paper gives our theoretical results for the approach of elimination of fields using equality edits, when present, to reduce the dimension of the edits; thus, the editing problem. It also presents comparative computational results of the C++ implementation with a SAS implementation that NASS developed earlier, also for the new system.
|