A Continuing Study on a New Resampling Method To Reduce Small-Sample Bias: An Extension to Multivariate Data

• Activity Number: 510
• Type: Contributed
• Date/Time: Thursday, August 2, 2007 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Computing
• Author(s): Haiyan Bai*+ and Wei Pan and Stephen Sivo
• Companies: University of Central Florida and University of Cincinnati and University of Central Florida
• Keywords: resampling ; bootstrap ; jackknife ; randomization tests ; cross-validation ; small-sample bias

Abstract

The four extant resampling methods-the bootstrap, and its counterparts: the jackknife, randomization tests, and cross-validation-are applicable to multivariate data, but with limitations of finding a single-valued statistic or combining the univariate p-values into a single p-value (Good, 2006). Recently, a new resampling method was developed to produce amplified samples with less bias, more robust to outliers, and larger statistical power than the pre-existing resampling methods; unfortunately, the new method is limited its application to univariate data. The current study extends the new method to multivariate data without the aforementioned limitations, providing a practical technique with a plug-in principle for researchers to employ advanced analysis directly on the amplified multivariate data with retained statistical properties, including correlations.