基于负二项分布的确切推断在微核试验数据分析中的模拟研究

A Simulation Study of Negative Binomial Exact Test for Small Sample Micronuclei Data

目的通过模拟研究对几种可用于小样本微核数据统计分析的方法进行讨论,为小样本微核数据的分析提供科学依据。方法用R软件进行编程模拟来估计不同参数下不同分析方法的Ⅰ类错误和检验功效。结果 Poisson精确概率法、Fisher精确概率法、负二项精确概率法、t检验及其变换、确切概率秩和检验的检验功效和Ⅰ型错误各不相同,其检验功效随δ和动物数n的增大而增大。其中Poisson精确概率法、Fisher精确概率法、负二项精确概率法的检验功效较高,但总体来自于负二项分布时,Poisson精确概率法、Fisher精确概率法的Ⅰ型错误较大,而负二项精确概率法Ⅰ型错误较小。结论在小样本微核数据分析中负二项精确概率法具有检验功效较高且犯Ⅰ型错误概率小的优越性。

Objective To explore the statistical analysis method of small-sample micronuclei data by simulation studies.Methods We simulated small-sample micronuclei data,then compared the size and power of the test methods in literature for the micronuclei assay analysis.Results Compared the typeⅠerror and statistical power of several different small-sample anaysis approading for micronuclei data,etatistical analysis methods,such as Poisson exact test,Fisher exact test,Negative Binomial Exact Test,t test,t test after transformation,rank sum exact test,were considered.The power of tests increase with δ and n.Statistical power of Poisson exact test was the highest.Fisher exact test and Negative Binomial Exact Test were lower than Poisson exact test,and higher than other statistical analysis methods.But under negative binomial distribution,type Ⅰ error was large in Poisson exact test and Fisher exact test,while it was small in Negative Binomial Exact Test.Conclusion The power of the Negative Binomial Exact was higher in spite of the lower type Ⅰ error.