Refinements to the Boolean approach to automatic data editing

Automatic data editing consists of three components: identification of erroneous records, identification of most likely erroneous fields within an erroneous record (fields to impute), and assignment of acceptable values to failing records. Moreover the types of data considered naturally fall into three categories: coded (categorical) data, continuous data, and mixed data (both coded and continuous). For the case of coded data, a natural way to approach automatic data is commonly referred to as the Boolean approach, first developed by Fellegi and Holt. For the fields to impute problem, central to the operation of the Fellegi-Holt approach is the explicit recognition of certain implied edits; Fellegi and Holt orginally required a complete set of edits, and their algorithm to generate this complete set has occasionally had the distinct disadvantage of failing to converge within reasonable time. The primary results of this paper is an algorithm that significantly prunes the Fellegi-Holt edit generation process, yet, nonetheless, generates a sufficient collection of implied edits adequate for the solution of the fields to impute problem. 3 figures