|
本帖最后由 Menuett 于 2013-12-22 15:59 编辑 9 A6 s% ]( q% y1 v% C
煮酒正熟 发表于 2013-12-20 12:05 ![]()
5 x/ @4 P: S. k7 p4 [$ |6 n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 ...
. `! p) }9 Z$ O5 T* H" l$ |9 F- j3 U1 s
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)。
+ g4 H9 J* w j1 n0 q4 e
% R( j% q' G5 W% P: V6 J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)5 G! c8 z3 [ N4 T) q' a- f0 W
3 I1 p7 s. N% k1 e. W# \; }4 E
R example:
. U0 H. n. o& V0 c6 G
8 T1 a. Y/ v" z5 @) `> M<-as.table(rbind(c(1668,5173),c(287,930)))
g5 b0 X% K. R) j9 n> chisq.test(M)
O, ]9 `9 h( C5 V8 Q$ c# v: c, ]; E/ t! X* ?
Pearson's Chi-squared test with Yates' continuity correction
6 E z* r3 t p4 K* u! u& C# J
5 O' v% u1 ~# M5 q4 r2 a% adata: M
( c( O3 H: ?9 v w' sX-squared = 0.3175, df = 1, p-value = 0.5731
( x9 e5 G' m5 ~" W N; P
. {# f; R* n5 j5 o# q# }Python example:3 R0 ~* U7 P! J! i3 T1 w9 F
8 t# ~, b6 n i% i3 L6 V: [* I>>> from scipy import stats9 P6 l# f* `/ Z: o' O8 E/ j
>>> stats.chi2_contingency([[6841-5173,5173],[1217-930,930]])& ? y4 N. l2 r/ X2 i" R
(0.31748297614660292, 0.57312422493552839, 1, array([[ 1659.73628692, 5181.26371308],
`# p1 _' E$ N [ 295.26371308, 921.73628692]])) |
|