一种随机化检验算法及其Matlab实现

在数据满足特定假设的前提下，可用有关统计技术检验样品间的差异．然而，在自然情况下，这些假设往往不成立或未知成立．本研究建立了一种随机化检验算法，可对实数域上的两个样品进行差异显著性分析，给出了Matlab标准源程序．算法具有广适性．设定随机化模拟次数和差异显著性水平，则可计算出检验值P，及差异显著与否．该算法可使用各种差异函数，如欧氏距离等等．应用本算法，对数种实测样品数据，以不同的差异函数进行了差异性分析．

An Algorithm for Randomization Test and Its Matlab Implementation

Traditional statistic methods could be used to test the difference when strict statistic assumptions on samples are met. However these assumptions are always not met in the nature. The statistic tests cannot be conducted when various difference measures are used. The randomization techniques bear less statistic restrictions. In this study we presented an algorithm for randomization test, which has a wider applicability. It may be used to test the statistic significance of difference between two samples with various indices. Times of randomizations and significance level should be set in the algorithm. The Matlab procedures of the algorithm were given. Difference function, e.g., Euclidean distance, should be defined in another Matlab file. Using the algorithm, difference analyses on human races and populations in China, and invertebrate diversity were made based on various difference functions, and some confirmative conclusions were drawn.