Statistical Commission and Conference of European Statisticians Unece Work Session on Statistical Data Editing Development of Modern Edit and Imputation Methods at Statistics Netherlands Contributed Paper Submitted by Statistics Netherlands

Topic (iv): Impact of new technologies on statistical data editing Abstract: The development of modern edit and imputation (E&I) methods and software is one of the spearheads of the Methods and Informatics Department of Statistics Netherlands. Many aspects of E&I are covered by the work that is currently being carried out. Software development focuses on the further development of SLICE, a general software framework for automatic E&I. At the moment, SLICE is being extended with the new Cherry Pie module for automatic error localisation in a mixture of categorical and continuous data, with a module for regression imputation of missing continuous data, and with the EC System module for modification of imputed values so that all user-specified edits become satisfied. Besides SLICE, the WAID program for donor imputation of missing values is being developed further. Finally, a software tool for graphical macro-editing based on the functionality offered by SPSS enhanced with Visual Basic modules that are integrated into the SPSS environment is being developed. Methodological research focuses on selective editing and automatic E&I. The merits of classification/regression trees and (logistic) regression for selective editing purposes are currently being investigated. For automatic E&I an ambitious research project is planned to start early 2002. This project will last four years. Aim of this research project is to develop an approach for automatic E&I that integrates the approach based on the Fellegi-Holt paradigm and the NIM approach. Like NIM our approach will use 'predicted' values to locate and correct errors. In contrast to the NIM, the approach will not be based on hot-deck donor imputation exclusively, but will also allow other types of imputation. If no imputation method is specified, the new approach will be the same as the approach based on the Fellegi-Holt paradigm. In this paper the above-mentioned development of E&I methods and software at Statistics Netherlands is described in some detail.