z***b 发帖数: 4667 | 1 【 以下文字转载自 Faculty 讨论区 】
发信人: zpytb (zpytb), 信区: Faculty
标 题: 请教一个统计问题
发信站: BBS 未名空间站 (Wed Oct 1 13:29:14 2014, 美东)
在做一篇文章,对比了5个东西(A,B,C,D,E),做的结果是
A=meanA+/-std1
B=meanB+/-std2
...
E=meanE+/-std5
reviewer说要提供confidence interval,大老板说那就算一下p value吧
我想到的是用student t test来两两比较谁大谁小,得到5x5的p矩阵,
我本来的意思是想从(A,B,...E)找出最好的一个方法 (mean越大越好),
但是结合这个p值的话就没有统一的趋势了,会出现
比如说 meanC>(meanA,meanB,meanD,meanE)
但是 p(C,A)<0.05, p(C,E)<0.05, 其他p>0.05
而且我还不知道要不要correct multiple comparison,到底是用
0.05,还是0.05/4,来作为threshold
请各位前辈指点一下,
有没有其他更好的统计量来衡量啊
大老板基本不回复任何学术问题了,谢谢 | m*******t 发帖数: 168 | 2 Yes. you need to correct multiple comparison if you want to compare all
pairs.
If you can pre-specify your research question, the multiple comparison can
be avoid. | c***z 发帖数: 6348 | 3 It is quite possible that A>B>C>D>E>A
You may need some experiment design help, which I am not an expert. | u*h 发帖数: 397 | 4 Maybe you can try Tukey range test ?
http://en.wikipedia.org/wiki/Tukey's_range_test
【在 z***b 的大作中提到】 : 【 以下文字转载自 Faculty 讨论区 】 : 发信人: zpytb (zpytb), 信区: Faculty : 标 题: 请教一个统计问题 : 发信站: BBS 未名空间站 (Wed Oct 1 13:29:14 2014, 美东) : 在做一篇文章,对比了5个东西(A,B,C,D,E),做的结果是 : A=meanA+/-std1 : B=meanB+/-std2 : ... : E=meanE+/-std5 : reviewer说要提供confidence interval,大老板说那就算一下p value吧
| c****y 发帖数: 94 | 5 It sounds like that you did a pairwise comparisons among those methods. You
need to apply adjustment. You actually should have 10 pvalue (5x4/2). If you
want to use Bonferroni Correction, it may be too conservative.In sas, you
can use proc multtest with hommel option (different way to control family
error rate). Hope it can work for you. If adjusted p-value is not
significant, it just means there are not significant difference between the
methods (2 methods compared). Actually, you should do ANOVA test at the
begining to decide whether there are significant difference among all
methods. Then if there is significant difference, you continue to do
pairwise comparison. If you can consult with a statistician,it will be the
best.
【在 z***b 的大作中提到】 : 【 以下文字转载自 Faculty 讨论区 】 : 发信人: zpytb (zpytb), 信区: Faculty : 标 题: 请教一个统计问题 : 发信站: BBS 未名空间站 (Wed Oct 1 13:29:14 2014, 美东) : 在做一篇文章,对比了5个东西(A,B,C,D,E),做的结果是 : A=meanA+/-std1 : B=meanB+/-std2 : ... : E=meanE+/-std5 : reviewer说要提供confidence interval,大老板说那就算一下p value吧
|
|