Transformation and Sample Size
Hui Jin, Xuejun Zhao, Richard Stridbeck · 2009
This thesis provide a discussion of the methods of reducing the sample size with xed power and signi cant level. Since many statistical tests are based on the assumption of normality, some transformations on the data should be done. We also focus on how large sample size will be needed and whether the sample size can be reduced after the transformations. We choose four kinds of distributions as an example. We apply three kinds of transformations and di¤erent statistical methods to compare the inuence on the skewness and sample size. The article rst illustrates the reason of doing the transformations on data. By investigating di¤erent transformations, the necessity of doing the transformations is motivated. And we also get the di¤erences on the reductions of sample size for the di¤erent distributions and transformations. Then we compare the di¤erences between two tests on original and transformed data. Eventually, we choose one of them and show it is a better way to deal with the data. key words: data transformation, sample size, skewness