|
|
本帖最后由 Menuett 于 2013-12-22 15:59 编辑 * ?9 H. w" m! l7 E
煮酒正熟 发表于 2013-12-20 12:05 8 r1 s5 S( N9 o5 g( \% Y* \
基本可以说是显著的。总的来说,在商界做统计学分析,95%信心水平是用得最多的,当95%上不显著时,都会去 ...
* _& E0 N6 c( D: N* M! ?2 o& A B: V' f$ Z
这个其实是一种binomial response,应该用Contigency Table或者Logisitic Regression(In case there are cofactors)来做。只记比率丢弃了Number of trial的信息(6841和1217个客户)。 " j. K: e7 r8 T! |
2 r* U. H2 n" J" t: I X
结果p=0.5731。 远远不显著。要在alpha level 0.05的水平上检验出76.42%和75.62%的区别,即使实验组和对照组各自样本大小相同,各自尚需44735个样本(At power level 80%)。see: Statistical Methods for Rates and Proportions by Joseph L. Fleiss (1981): Z% j6 H! I0 ~+ p( l
4 a+ ?3 L7 p) n J7 i& ?
R example:! D$ J. o+ o. E7 V; q. a
$ L9 n( }6 n B, y: T> M<-as.table(rbind(c(1668,5173),c(287,930)))
& | z* l0 C7 N5 S% D> chisq.test(M)
1 k$ ~+ R( B8 A; ]3 D
) K/ R8 V K. I, [8 C9 H# K Z Pearson's Chi-squared test with Yates' continuity correction7 V( u4 j; w3 K7 E G# N
+ ~+ Q- j' T1 [3 Bdata: M
* R& [# v. H3 u5 w- M8 LX-squared = 0.3175, df = 1, p-value = 0.5731
3 @' j( e' M0 t
5 l; Y* `, P) O& W+ t5 nPython example:) r/ q A% C+ c# y- Q S; B0 L
0 _. _$ H% |3 d1 l>>> from scipy import stats G; Y/ D& L8 {- [
>>> stats.chi2_contingency([[6841-5173,5173],[1217-930,930]])/ R, I* N% S6 H" w( m x2 ?
(0.31748297614660292, 0.57312422493552839, 1, array([[ 1659.73628692, 5181.26371308],, \& B, ~6 Q, {9 Z# |5 M
[ 295.26371308, 921.73628692]])) |
|