几种常见的显著性检验方法
1.Tukey (John Wilder Tukey) test德国哲学家
梦见棉被最著名的有2个:
(1)Tukey test for multiple comparisons
主要应用于3组或以上的多重比较。比如说一共有4组数据,两两比较产生6个统计值,Tukey-test用于生成一个critical value来控制总体误差(Family wi error rate,FER),与Tukey test相类似的是Dunnett test,它是控制多对一比较(即3组同时和一个参照组比较)的FER。
(2)Tukey trend test
红楼梦的主要人物主要用于检验同一药物不同剂量下和参照药物的线性关系。Tukey trend test 简单但及其高效,是生物统计学常用的方法。
2.T-test
最大的哺乳动物
土家族文化T检验,这是1905年st氏首先提出的,当时他以“Student”为笔名发表,故至今有的书籍仍称之为“学生氏检验”。t可能是倍数的意思(times),t就是样本均数SX(x)与总体均数(“)间相距几倍标准误(sx)。t检验是用于比较两均数间相差是否显著的。
t检验过程:是对两样本均数(mean)差别的显著性进行检验。唯t检验须知道两个总体的方差(Variances)是否相等;t检验值的计算会因方差是否相等而有所不同。也就是说,t检验须视乎方差齐性(Equality of Variances)结果。所以,SPSS 在进行t-test for Equality of Means的同时,也要做Levene's Test for Equality of Variances 。
3.Dunn’s multiple comparison test王者荣耀赠送皮肤
Dunn's test calculates a P value for each pair of columns. The P values answer this question: If the data were sampled from populations with the same median, what is the chance that one or more pairs of columns would have medians as far apart as obrved here? If the P value is low, you'll conclude that the difference is statistically significant. The calculation of the P value takes into account the number of comparisons you are making. If the null hypothesis is true (all data are sampled from populations with identical
distributions, so all differences between groups are due to random sampling), then there is a 5% chance that at least one of the post tests will have P<0.05. The 5% chance does not apply to EACH comparison but rather to the ENTIRE family of comparisons.
Dunn's test compares the difference in the sum of ranks between two columns with the expected average difference (bad on the number of groups and their size). For each pair of columns, In Stat reports the P value as >0.05, <0.05, <0.01 or < 0.001. The calculation of the P value takes into account the number of comparisons you are making. If the null hypothesis is true (all data are sampled from populations with identical distributions, so all differences between groups are due to random sampling), then there is a 5% chance that at least one of the post tests will have P<0.05. The 5% chance does not apply to EACH comparison but rather to the ENTIRE family of comparisons.
>九七年的今年多大