|
本帖最后由 Menuett 于 2013-12-22 15:59 编辑
9 H& E- z q' ?, A2 M. q) j8 @煮酒正熟 发表于 2013-12-20 12:05 ![]()
2 R" _9 n1 T8 t* @8 Z/ b6 ?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 ...
9 @. R/ X7 y/ n6 |6 B5 n" m& G8 t4 u0 H5 ^& [6 U
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- t. n# F ?# k, S% {7 k' d5 j! v9 ^9 }
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)
6 W% X' R& U6 u: f } Z4 _& p
4 H0 H" p) T# M Q( L3 {R example:
; e+ N) } R. E3 K) a) F2 p" B1 V2 l
> M<-as.table(rbind(c(1668,5173),c(287,930)))
- D- ]0 `1 f: M> chisq.test(M)
v" ]) q4 R2 R; T
, ?* n* N; I" C+ j1 q6 F8 o Pearson's Chi-squared test with Yates' continuity correction
. h4 v+ X) y. k' I) _& p5 Z2 t7 `: L& W: B; g# Z
data: M+ i% Y9 ^" {0 i6 f! x5 ^
X-squared = 0.3175, df = 1, p-value = 0.57316 ?1 ?% o9 B( y6 V& C
J: D$ D. I. {
Python example:
7 F) S- a3 \+ G2 P
, _" U7 |, d6 B>>> from scipy import stats% \. y' Z! c- E# p
>>> stats.chi2_contingency([[6841-5173,5173],[1217-930,930]])
" N% w7 X7 c0 K4 C5 d( L N(0.31748297614660292, 0.57312422493552839, 1, array([[ 1659.73628692, 5181.26371308],
5 {; Q4 @, \3 H( G5 k$ `& w [ 295.26371308, 921.73628692]])) |
|