Setting M-Estimation Parameters for Detection and Treatment of Influential Values

• Activity Number: 365
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Survey Research Methods Section
• Author(s): Mary Mulry*+ and Broderick Oliver and Stephen Kaputa
• Companies: Federal Employee and U.S. Census Bureau and U.S. Census Bureau
• Keywords: outlier ; economic surveys

Abstract

Recent research on the use of M-estimation methodology for detecting and treating verified influential values in economic surveys found that initial parameter settings affected its effectiveness. The study relied on simulated data designed to reflect the population properties for two industries in the Monthly Retail Trade Survey (MRTS), but the approach to determining settings for the initial parameters used an empirical analysis that does not generalize well. The need to expand the application of the M-estimation methodology to all the industries in the MRTS and the Monthly Wholesale Trade Survey stimulated the development of a more general methodology that uses historical data to determine the initial parameter settings. This paper discusses the effectiveness of several methods for setting initial parameters under several scenarios for the occurrence of influential values.