Weighting Adjustment Method

• Main Authors: Shih-Keng Chen, 陳世耿
• Other Authors: Zu-Wei Zheng
• Format: Others
• Language: zh-TW
• Published: 2004
• Online Access: http://ndltd.ncl.edu.tw/handle/7ack8j

碩士 === 中原大學 === 應用數學研究所 === 92 === The classical inference method for opinion survey is generally inappropriate when certain amount of the subjects sampled refused or were unable to provide responses. Based on the assumption that the missing mechanism is at random, differential weights derived from poststratification, which is closely related to the Horvitz-Thompson estimator, can be used to modify the estimation to adjust for bias. It is shown that under the missing-at-random assumption the estimator is unbiased. The method is primarily used to handle unit nonresponse, where a subset of sampled individuals do not complete the survey because of noncontact, refusal, or some other reason. We consider the applications to three common sampling schemes, namely, simple random sampling, stratified random sampling and cluster sampling. Monte-Carlo simulations show that the weighting adjustments have significantly improved the accuracy of the estimators in many cases.