Comparison of GEIS and SPEER for Editing Economic Data

Two computer systems are currently available for editing continuous, economic data: Statistics Canada's General Edit and Imputation System (GEIS) and the Census Bureau's Structured Programs for Economic Editing and Referrals (SPEER). GEIS, the more general of the two systems, uses linear inequality edits and provides several imputation options. SPEER uses ratio edits which are a special case of linear inequality edits and provides only one imputation option. The main algorithm of GEIS is a relatively fast extension of Chernikova's algorithm for finding extreme points in n-dimensional space. SPEER edits are exceedingly fast because they do not involve advanced integer programming algorithms. The methods of the two systems are described and an empirical comparison is provided. 1. INTRODUCTION Computer files used for administrative or survey purposes may contain large numbers of records, some of which contain logical inconsistencies or incorrect data. Pritzker, Ogus, and Hansen (1965) describe the nature of the problem. Errors can arise because methods of creating records in files are not consistent, because questions are not understood, or because of transcription or coding problems. In many situations, data files are edited using custom software that incorporate rules developed by subject-matter specialists. If the specialists are unable to develop the full logic needed for the edit rules, then the subsequent edit software is in error. If programmers do not properly code the rules, then the software would be in error. Developing software from scratch each time a data base is redesigned is time-consuming and error-prone. It is better to have a system that can describe edit rules in tables that are read and utilized by reusable software modules. The tables could be more easily updated and maintained than complex if-then-else rules in computer code. The software would automatically check the logical validity of the entire system prior to the receipt of data during production processing. Fellegi and Holt (1976), hereafter FH, provided the theoretical basis of such a system. FH had three goals that we paraphrase: