Total Survey Error in Practice

15Quantifying Measurement Errors in Partially Edited Business Survey Data

Thomas Laitila,^1,2 Karin Lindgren,³ Anders Norberg,³ and Can Tongur³

¹ Department of Research and Development, Statistics Sweden, Örebro, Sweden

² Department of Statistics, Örebro University School of Business, Örebro, Sweden

³ Process Department, Statistics Sweden, Stockholm, Sweden

15.1 Introduction

To ensure high quality in a business survey, a statistical institute often uses an editing process to detect any inconsistencies and other errors in the data collected. The process of identifying errors and suspicious values and of manually reviewing and revising data is costly and time consuming, because data revision often involves recontacting the respondents. Development of new theories and methods is of interest as it may result in substantial cost savings and more timely production of statistics. One way to find potential improvements is to develop more efficient tools for identifying erroneous observations. Another is to reduce the number of observations manually reviewed, the aim of selective editing (SE). Indeed the traditional approach of correcting all errors is not generally necessary for appropriate statistical inference (Granquist and Kovar, 1997).

SE is defined by Granquist and Kovar (1997) as editing methods that only select a subset of the responses that failed edit rules for manual review. Here the leading idea is to expend resources only on those suspected data that may have major effects ...