|
本帖最后由 Menuett 于 2013-12-22 15:59 编辑
6 c) r# M5 s& |3 ]0 l煮酒正熟 发表于 2013-12-20 12:05 - Z3 Q& ]6 b0 ]: ?0 ^
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 ... : j2 N: W* T u
2 ?" j1 @: |2 a9 X7 M+ b* ]& d# c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)。 # V$ L% K! A6 K/ c
7 Z/ B' i$ f* f0 z @) e, I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)" e: ` @* U( x: M& V
& l: Q) q D0 _. B' V# C5 c8 a
R example:
) `# V Q, E6 l/ o0 }9 A" `+ O
' k: u0 g% x6 r> M<-as.table(rbind(c(1668,5173),c(287,930)))
0 B/ B4 E. f6 U$ j/ ?> chisq.test(M)$ D) B; {, F4 M6 C
+ J+ `; K7 Z0 Z: X' b! T Pearson's Chi-squared test with Yates' continuity correction
7 _ @) ]' C& ]2 O9 ~
1 ]0 d1 D3 x* m! E" {data: M. j3 t) b" y e. i
X-squared = 0.3175, df = 1, p-value = 0.5731
, m; U4 X% U$ D0 D* j- O$ _6 [0 j2 ~) j2 D j+ J- k. g
Python example: i5 k8 w; x/ e1 o, L) e2 o$ u. \
, p6 D* ~9 r1 q5 P! g
>>> from scipy import stats4 j7 I, }/ _: ]
>>> stats.chi2_contingency([[6841-5173,5173],[1217-930,930]])* Q4 ]7 P9 l& j# J. X0 W H
(0.31748297614660292, 0.57312422493552839, 1, array([[ 1659.73628692, 5181.26371308],' Y# H* I( u+ V0 R& O
[ 295.26371308, 921.73628692]])) |
|