Use of Tax Data for Replacing Business Survey Data - The Data Quality Issue

Using various administrative and tax data sources to replace key variables collected in business surveys can significantly reduce both survey cost and response burden. Data collected from administrative sources, however, are generally not subjected to the same data quality and consistency checks as those for survey data, and often do not correspond exactly with the variables targeted in the survey questionnaire. Successful use of administrative sources to replace survey data depends to a certain extent on how effectively we can address the issues of ensuring inherent data quality and adequate correspondence with the survey variables. In this work, we present some preliminary results on using multivariate methods, Mahalanobis distance in particular, for this purpose in business surveys. Our emphasis is on the examination and utilization of the internal relationships and structures of the tax data themselves for identifying records that have large differences between the two sources and that should be treated in producing business statistics. Our results demonstrate the utility of multivariate methods in at least three key areas: direct replacement of survey response by tax data, modelling of survey response by tax data, and identification of misclassified tax records.