|
本帖最后由 Menuett 于 2013-12-22 15:59 编辑
/ J6 w! ~1 V' m% V1 f煮酒正熟 发表于 2013-12-20 12:05 2 E/ I0 r D7 ?4 d7 m1 I
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 ...
& P& O5 b" K2 \+ G0 [6 ?1 x% c9 o, h! [
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)。 # D' s3 I& W* G' X4 Q( u! l
; E0 l2 S- o% ]9 E3 V
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); T) [1 C. e3 m! J) i( D8 I
2 U! _) |& q( ^8 W5 \% J; `R example:
) P, g& u2 j* X3 n8 Q! U% Q5 X4 O. d- G7 k0 G; X2 v
> M<-as.table(rbind(c(1668,5173),c(287,930)))
3 Z( P6 ^; h. R, k3 R> chisq.test(M)
0 w8 o% H3 Q2 f, [! T) E- s2 s1 N7 v# {* s; q# |9 j& |& H. k8 R' }
Pearson's Chi-squared test with Yates' continuity correction: S% m- C: g1 T! B% C: a
$ B: t/ Y5 ` m! e# M
data: M! Q! p' s& |& B$ ^( M2 m# g
X-squared = 0.3175, df = 1, p-value = 0.57318 f: d. L" S4 r0 l- W
0 q1 x3 n- e: C# X$ u V
Python example:
+ }1 r# \' p! S: J+ H3 U6 ^4 ^7 R, E: T3 [! B4 {( z/ h( {5 \* h
>>> from scipy import stats4 m8 Q& [% [9 R$ A! _; Q, h: ~
>>> stats.chi2_contingency([[6841-5173,5173],[1217-930,930]])
* _$ b1 L$ m$ j( ?+ u3 e/ _(0.31748297614660292, 0.57312422493552839, 1, array([[ 1659.73628692, 5181.26371308],
. n9 X' t6 @, Y7 b- z [ 295.26371308, 921.73628692]])) |
|