|
本帖最后由 Menuett 于 2013-12-22 15:59 编辑
+ i, z1 ~* m9 I6 j# b- q# l( \+ t煮酒正熟 发表于 2013-12-20 12:05 6 {2 C* v% A+ j8 i- C3 b2 { H B
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 ...
3 I/ W, b, ^8 D; Z6 T/ P4 C4 e) t' o' Q$ ~ z/ D/ }
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)。 $ ^' [+ O: k1 k* H9 `# J4 Z
# Y/ Y0 e3 N7 J) M* S$ S
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)
2 [7 m& k1 {& L) U( R/ A( @& t# `% T5 r ]2 ?$ _
R example:
* d6 S. _6 V" |7 a" E- A% K
+ U& d) W6 X0 f' M> M<-as.table(rbind(c(1668,5173),c(287,930))); h2 |; b8 ^5 E8 f/ r
> chisq.test(M)# [0 B5 Y* p. Q2 m$ G, l- ^
# `1 h3 F9 f/ D" ~ Pearson's Chi-squared test with Yates' continuity correction- o, \% X3 G4 |, e4 A7 F
}% @, F3 y6 O# o. X- I3 \, ]
data: M
8 h: ]$ F( G% z, uX-squared = 0.3175, df = 1, p-value = 0.5731
5 k7 O! ?0 l: h4 ]
0 p* t% A) x5 G. VPython example:
4 J- G0 p: w' x! W% S' t/ y; Q/ c9 r. O' A
>>> from scipy import stats
/ `4 l- l4 |8 i6 N>>> stats.chi2_contingency([[6841-5173,5173],[1217-930,930]])& o& p5 z; |9 a0 t8 R& ^, p- m
(0.31748297614660292, 0.57312422493552839, 1, array([[ 1659.73628692, 5181.26371308],
: b3 x4 {4 O( q* A. e, ] [ 295.26371308, 921.73628692]])) |
|