|
本帖最后由 Menuett 于 2013-12-22 15:59 编辑
- T |* r( d! T2 X# M6 q0 p煮酒正熟 发表于 2013-12-20 12:05
6 }9 X5 T" s- B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 ...
, C/ G" {) m$ Q6 _% U) _$ ~3 O. L* j$ a3 u" V0 m8 X: X
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)。
& m4 Z& D9 J; R y# H3 q8 z4 g: l T3 _) ^4 ^
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)# U# Z2 A' |& C4 Q0 g* R
5 N: M3 t$ j7 q( r8 C9 q( _R example:* c8 O) G2 A, h8 N
2 O+ _! A" M' j* x' G: ?: c; o
> M<-as.table(rbind(c(1668,5173),c(287,930)))
0 H9 l' `" T8 J3 t0 t6 }> chisq.test(M)
9 N: T' v# L f. H' v1 b; m) W3 @4 b0 f6 v3 K( x/ E2 g4 b' H1 P
Pearson's Chi-squared test with Yates' continuity correction' d( H) C3 G. S$ Q! g1 b+ P
8 F5 ^8 h2 _9 d! Y7 a4 r
data: M, B& g1 F" S7 F g7 j3 J8 e" ^
X-squared = 0.3175, df = 1, p-value = 0.5731
7 Z1 O$ Q& l# O+ x- X+ _
- F( O3 s7 }! f2 M* I) ]* ^* qPython example:
5 d# N' ?" u* y, h# P3 J% v$ S
3 D* f% z( V. z* j5 D>>> from scipy import stats O( z- P! y3 @( D7 f
>>> stats.chi2_contingency([[6841-5173,5173],[1217-930,930]])5 R' E) P t+ Q
(0.31748297614660292, 0.57312422493552839, 1, array([[ 1659.73628692, 5181.26371308],
7 }% z" q! d9 Q9 F9 Q [ 295.26371308, 921.73628692]])) |
|