|
本帖最后由 Menuett 于 2013-12-22 15:59 编辑
; _. ?+ E1 o- O( N, U0 w! h5 G9 v1 A煮酒正熟 发表于 2013-12-20 12:05 9 k' L% v4 ^0 U1 W
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 ... + q/ ~" _1 Y# y
; ^# }8 A2 i: [9 o# Z% q4 q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* t5 ?+ N& r. S9 E
9 @* b* j. M* ~ B0 J7 S5 f9 u4 m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) w i( m; o/ I& S! }. L
8 n5 L* A5 ^ |
R example:
+ m/ m1 N5 u& ?( [. V% t5 F/ e6 Z4 D. @
> M<-as.table(rbind(c(1668,5173),c(287,930)))
; [, }7 W3 D" Q/ N7 L> chisq.test(M)
* r5 P& _/ v) ^( h. k/ F9 z3 _! R; g
Pearson's Chi-squared test with Yates' continuity correction
$ N0 y$ R, \: `% ~' \1 T9 B8 e/ |( H+ {! K+ Z0 K
data: M
# M6 I% M. X9 x# UX-squared = 0.3175, df = 1, p-value = 0.5731" G% v( x! }& P2 K" z" U# P
' D; W9 c. a8 j4 F4 HPython example:. D8 E1 O) t/ W2 i8 }/ G7 P
& s' W% R" V2 l, u7 j1 T
>>> from scipy import stats) Z$ a8 m, i6 o2 o/ S" z/ Y; f
>>> stats.chi2_contingency([[6841-5173,5173],[1217-930,930]])
7 S5 A! G3 N$ {+ @(0.31748297614660292, 0.57312422493552839, 1, array([[ 1659.73628692, 5181.26371308],
3 U$ e: O2 z( b+ |8 ~" r; c [ 295.26371308, 921.73628692]])) |
|