Detection of Unlikely Answers in Survey Research

In recent years a lot of attention has been dedicated to non-sampling survey errors, and many editing procedures have been elaborated. The widespread use of computer-assisted surveys makes it possible to perform on-line checks and corrections, which are particularly useful, because the respondent himself can solve the eventually detected inconsistencies. Most of the editing procedures are limited to detect and correct impossible answers or combinations of answers. This contribution investigates the feasibility of a general automatic on-line procedure to detect not only impossible, but also inconsistent and unlikely answers.