|
本帖最后由 Menuett 于 2013-12-22 15:59 编辑 : \5 |8 i9 G6 D% p# k& v2 f
煮酒正熟 发表于 2013-12-20 12:05 ![]()
0 ?. o) B0 U" |7 p$ e) o6 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 ... : v+ \: S9 x: W( S# R: y9 y
9 J+ H; y) s b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)。
5 o% l; x4 D Q: U& u4 j5 o; W, [. K* m& G" P
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 U/ \3 p$ T% ]5 g2 K8 I; F' {& r
! S4 v& A" A0 ]: u* t3 S, z6 fR example:
, L4 r- k; ~5 r9 ]7 [3 X4 }! q, {$ B1 E" Q* f/ E* F9 X: t
> M<-as.table(rbind(c(1668,5173),c(287,930)))2 U3 v! b# }1 w7 h! t% `
> chisq.test(M)6 N0 ]# V6 U9 _, A7 f+ ?2 Y
- d3 t8 O# G$ [7 g; h; u Pearson's Chi-squared test with Yates' continuity correction, f# `, m S: w& P% @6 P6 n; H
' N# ]0 h5 K: q: sdata: M
/ y; @4 e* O; jX-squared = 0.3175, df = 1, p-value = 0.5731, | C" {8 c: q4 V7 w8 f
6 w5 z, `- @5 {/ `7 WPython example:, Q/ N# @, P/ y% k. M
1 [' q$ }0 e! H>>> from scipy import stats
1 A% m& a$ _- A5 D; ~ r5 H1 a1 |>>> stats.chi2_contingency([[6841-5173,5173],[1217-930,930]])
: ^9 H! l8 }" K; m8 k1 V( `- g- |2 o(0.31748297614660292, 0.57312422493552839, 1, array([[ 1659.73628692, 5181.26371308],
/ L% E7 X1 o* \/ V2 [0 G: b' [' ?. u [ 295.26371308, 921.73628692]])) |
|