|
本帖最后由 Menuett 于 2013-12-22 15:59 编辑
+ u" V( f) G: y; _( T/ q煮酒正熟 发表于 2013-12-20 12:05 6 n9 K9 b9 k# C
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 ... 8 i7 W2 `* c8 I4 }9 H
) {' |5 T2 X" K2 z- U9 ?0 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)。
2 K2 G$ ^3 n' ] B7 K6 j! M& Z- L/ E' s, T4 x
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)% C/ B5 O) j: J3 W' b. i2 H
0 l* y- T# Y/ r$ R3 {' g6 BR example:
$ c2 U% `2 @$ [$ |: r" U
$ k+ _9 z% j% t& H& m> M<-as.table(rbind(c(1668,5173),c(287,930)))
( T4 k. p* e0 \2 J6 v4 V> chisq.test(M)4 m4 e5 H& g9 l4 K( H- G7 @# W! {
! R: t" [" A! e1 f! O Pearson's Chi-squared test with Yates' continuity correction. o J9 T3 S! n1 a! b
5 t; w/ c6 |, k& [
data: M
y3 k, W$ U. z7 |6 S) }X-squared = 0.3175, df = 1, p-value = 0.5731
; r% Z# w6 J- P5 G% ^- }
: k$ p8 v; @' `; _ [6 r7 r9 YPython example:2 K4 ~: s+ {+ W1 ~/ a9 S
K) B* q2 S# i- G$ _>>> from scipy import stats
8 n. I! n' j: C4 Y9 c5 N>>> stats.chi2_contingency([[6841-5173,5173],[1217-930,930]])
) E9 u) w+ ?8 U$ w8 I& M(0.31748297614660292, 0.57312422493552839, 1, array([[ 1659.73628692, 5181.26371308],
9 B! z; `, u2 S ]2 i4 O& C [ 295.26371308, 921.73628692]])) |
|