|
|
本帖最后由 Menuett 于 2013-12-22 15:59 编辑 + l: ?. H$ S: ?/ _" ^4 K0 q
煮酒正熟 发表于 2013-12-20 12:05 9 H# k% Q/ @9 D
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 x, P2 f' h7 g7 D+ |9 p* b
1 o# {6 m, ]; G1 l" 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)。
, ^7 b+ h* q0 x7 P' y' y) j, i4 @& o
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). E- B2 W. s, \$ @" F% d' F% M
* s& W' A" ?7 D3 W+ s+ p9 Z- e
R example:2 `; z2 L+ s- a+ o- o9 a, S
. ~, t, m5 L9 x0 |$ T0 r
> M<-as.table(rbind(c(1668,5173),c(287,930))), G$ h( L+ a/ Y9 B* I
> chisq.test(M)* ^. h# u1 n( K& P1 J6 _; B
& C# p2 N2 w( O6 r Pearson's Chi-squared test with Yates' continuity correction
# D9 l: x- y; I1 W5 P
7 O) H; B# E1 R1 T. j/ Edata: M
) u* V& O8 }- n- r% W+ L% RX-squared = 0.3175, df = 1, p-value = 0.5731
& \# n( I. Z% e! ~2 o
5 K( _# u& o' j! bPython example:% o# R! A$ w8 Y7 K5 m' l6 }
# o0 a% {3 q7 T6 h0 \$ z8 U
>>> from scipy import stats/ r5 Z6 z: `1 h9 }% B0 f
>>> stats.chi2_contingency([[6841-5173,5173],[1217-930,930]]): a/ _6 T0 n0 i x5 W3 K- Y3 l/ b
(0.31748297614660292, 0.57312422493552839, 1, array([[ 1659.73628692, 5181.26371308],
( T0 L5 u( a; G [ 295.26371308, 921.73628692]])) |
|