|
|
本帖最后由 Menuett 于 2013-12-22 15:59 编辑
5 T$ F+ T) d) b, a9 D9 v1 L煮酒正熟 发表于 2013-12-20 12:05 % _3 U% Z+ b2 e& u6 w
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 ...
S; ^! Q. U1 B4 L5 }( R6 z/ T( K- m# |1 S* f5 Q5 \; b4 O2 }2 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)。 - P. ?6 B" ~, R$ z
1 g' s$ h3 [3 V& D* f3 O I* p0 M
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)
% c2 K5 q. l0 u
( K& g9 \" k+ \% X5 SR example:
+ W- h0 k4 P& v ~/ T$ x
. A. k; }% j( t& n6 i( Z) g> M<-as.table(rbind(c(1668,5173),c(287,930)))! h+ S" a3 @+ }! g6 }
> chisq.test(M)
7 {, k3 ^% E/ V' }
2 i2 |. }5 R) k& q+ j, } Pearson's Chi-squared test with Yates' continuity correction* I% o% ?8 l1 I& v0 p) F& n
! a! ~8 Z" i, G+ E/ N
data: M
& l; y& Z1 V' h6 `) j4 t. cX-squared = 0.3175, df = 1, p-value = 0.57312 A. g$ r( S4 m, r8 R9 @* V1 z, o
$ N: c0 G3 ~" k! d- K: Q; Z( nPython example:! Y! G9 v7 G+ h% \) `
$ x; g, E$ x. N$ J>>> from scipy import stats
5 m$ V4 ~" T9 G4 N>>> stats.chi2_contingency([[6841-5173,5173],[1217-930,930]])
( a* q5 M$ Z7 p6 S(0.31748297614660292, 0.57312422493552839, 1, array([[ 1659.73628692, 5181.26371308],* x2 m6 M' T4 p$ ^/ d% O, z P) U2 G
[ 295.26371308, 921.73628692]])) |
|