my3966
我现在要比较两组数据之间是不是有统计学上显著性差异,我本来要用独立数据的成组t检验的,但发现两组数据都不符合正态分布,我想用非参数检验,但是我的数据n=2000,我发现wilcoxon秩和检验,好像不能用于数据这么多的数据。
请各位大侠给我点建议,不胜感激!!!
yihui
不要“好像”不能用于这么多数据,谁告诉你Wilcoxon秩和检验不能用于大样本的?
neige
when your sample size is large, apply central limit thoerm, forget about the wilcoxon, i guess you can go ahead with T test
my3966
我用Wilcoxon秩和检验中大样本时的U test(alpha=0.01), 但结果根本不对,很明显的没有差别,而统计结果却是差异很显著。所以,在大样本时Wilcoxon秩和检验应该很不准确。
I have tried t-test and the results are quite different from commom t-test results. I took some data with no difference at all as a check and the t-value was always larger than 10(n=2000). So I am not very confident with the result of t-test.
yihui
你说的“很明显的没有差异”这个结论怎么得出来的?肉眼观察?
我建议你先对两组数据作箱线图比较。
drewlee
[quote]引用第2楼neige于2007-05-14 08:23发表的“”:
when your sample size is large, apply central limit thoerm, forget about the wilcoxon, i guess you can go ahead with T test[/quote]
How large can be so-called large?
neige
do not worry about the cut point, 2000 is way big enough,
drewlee
I don't think the figure 2000 is a magic number that is sufficiently big for every circumstance.
Another uncertainty is that we don't know whether my3966's H0 is that no difference between means of two samples.
neige
"我现在要比较两组数据之间是不是有统计学上显著性差异"
what else can it be? of course 2000 is not sufficiently big for EVERY circumstance, nothing is absolute. Do not be picky here
drewlee
How about the difference of two distributions?
y.luo
呵呵 还是麻烦my3966把问题说得清楚一些好
不然什么问题都不清楚,怎么解决啊?
比如,到底是独立样本还是配对样本?不过没有见过哪个统计方法说样本太大了不能用得