A Generalization of the Coefficient of Variation with Application to Suppression of Imprecise Estimates

• Activity Number: 229
• Type: Contributed
• Date/Time: Tuesday, August 10, 2004 : 10:30 AM to 12:20 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Avinash Singh*+ and Matthew Westlake and Moshe Feder
• Companies: RTI International and RTI International and RTI International
• Keywords: effective sample size ; exact binomial ; length of confidence interval ; length of discrimination interval

Abstract

We propose a generalization of the usual coefficient of variation (CV) to address some of the known problems when used in criteria developed to determine the suppression of estimates. Some of the problems associated with the CV in this application include interpretation when the estimate is near zero, and the inconsistency in the interpretation about precision when computed for different one-to-one monotonic transformations. The proposed measure, termed discrimination coefficient of variation (DCV), generalizes the CV using hypothesis-testing ideas involving length of the confidence interval (LCI) and the length of a discrimination interval (LDI). This discrimination interval is used to check whether the sample is large enough or the confidence interval is short enough to discriminate with certain power between the current value and postulated extreme values on either side of the current value. DCV allows for using an exact distribution for computing LCI such as the use of exact binomial when the estimated proportion is very small. The method is illustrated using data from the 2002 NSDUH (National Survey on Drug Use and Health) sponsored by SAMHSA.