A New Resampling Method To Reduce Small Sample Bias

• Activity Number: 30
• Type: Contributed
• Date/Time: Sunday, August 7, 2005 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Computing
• Author(s): Haiyan Bai*+ and Wei Pan and LihShing Wang
• Companies: University of Cincinnati and University of Cincinnati and University of Cincinnati
• Keywords: resampling ; small sample bias ; simulation

Abstract

Resampling as a revolutionary methodology to deal with small sample problems emerged with the development of modern computer techniques. Randomization exact test, crossvalidation, jackknife, and bootstrap are the four existing major resampling methods. However, they have inevitable limitations, such as lack of independent observations and little robust to the influence of outliers. This study attempts to reduce the limitations of the existing resampling methods by developing a new resampling method to obtain an enlarged, less biased sample from a small random sample, based on the unbiased confidence interval estimation for the mean. The enlarged sample is a union of randomly generated multiple samples, each with a mean determined by each of the cut-off points that equally divide the confidence interval of the original small sample and with the standard deviation of the small sample. The enlarged, less biased sample has larger statistical power, independent observations, and less influence of outliers. The new resampling method will be illustrated through a simulation study and compared with the existing resampling methods using empirical data.