|
本帖最后由 Menuett 于 2013-12-22 15:59 编辑 - w( o# v5 X, C1 R9 I+ V
煮酒正熟 发表于 2013-12-20 12:05 / t$ `0 q6 c4 B1 a
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 ... ) g% n6 {7 Q/ Y9 H" c
3 s/ e7 z+ H7 [% z+ w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)。 , ~7 e* ~' H0 x
. k' N0 ?: f: k+ ]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)1 s, s' h/ q0 c: D0 g2 W, b
, R$ ?5 `8 l; z9 y5 b6 T s
R example:
# i& ?: z2 F' |
4 B2 }8 z9 w# G! P> M<-as.table(rbind(c(1668,5173),c(287,930))) \) l1 \& t% y" A9 A- s! V! \
> chisq.test(M)
" r# M" A/ `' d. ^+ \8 [
/ V+ D5 h# p4 ?% M Pearson's Chi-squared test with Yates' continuity correction9 b- G( k0 o' Y; q5 z4 G
) }2 X, B/ a# \data: M
# _; J$ c' W0 k% E. h' m4 x, yX-squared = 0.3175, df = 1, p-value = 0.5731
m. |$ T% V, t9 Z, d: d* d1 _( P8 F8 h. ]& H
Python example:
4 B1 B5 x# p( ?2 b$ t
3 }; s" J8 Z( X R1 u$ {0 g>>> from scipy import stats
# P7 T" t7 N8 w j( Q>>> stats.chi2_contingency([[6841-5173,5173],[1217-930,930]])3 O- q" [) T ~. {5 D
(0.31748297614660292, 0.57312422493552839, 1, array([[ 1659.73628692, 5181.26371308],
8 {6 i! ]9 f0 P2 {/ q" d, n/ Q [ 295.26371308, 921.73628692]])) |
|