|
|
本帖最后由 Menuett 于 2013-12-22 15:59 编辑
. e9 S$ g6 a& [ {' z$ R9 C煮酒正熟 发表于 2013-12-20 12:05 6 r! B" y$ Q% R5 V9 |( 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 ...
' J% a9 y5 i8 q; X( ^6 s1 d! }% \! h
1 O4 D) C: N: @# U) r4 S) Q8 F5 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)。
0 d0 }: S; X7 a9 T3 H o; M$ Y, m8 x4 u" T, C) u
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)
/ h+ H3 J' y4 n" I0 j" N
/ e2 `: S; h) {" ^R example:
) ?1 Q! y3 R$ L0 F. q4 L; B; F5 G% {9 w$ t5 c0 l: R! ~& v& Z; n
> M<-as.table(rbind(c(1668,5173),c(287,930)))9 d2 c5 ?# B% v( D5 S
> chisq.test(M)( w( q% H( O$ x. {
! e7 ^1 @& ^: e6 j8 `6 \0 J3 B+ k) ^ Pearson's Chi-squared test with Yates' continuity correction
6 y# ]" o( K$ w6 A- @& A
) l3 v1 k7 y% \& C4 b4 |- E; ddata: M
% l. c1 E9 w8 k5 hX-squared = 0.3175, df = 1, p-value = 0.5731
: G9 G# h+ S/ Q; }, |! G7 n: u. ?# i# q4 N3 `
Python example:0 c4 |- d x N+ R8 r
2 k4 B% X6 S4 j( S0 t) r g
>>> from scipy import stats Y. Z8 P6 x$ z: w1 j/ i+ K
>>> stats.chi2_contingency([[6841-5173,5173],[1217-930,930]])+ q+ a! ^4 ^: p
(0.31748297614660292, 0.57312422493552839, 1, array([[ 1659.73628692, 5181.26371308],
, P' o4 p' _' @$ O z8 l4 y9 u! w2 s5 b [ 295.26371308, 921.73628692]])) |
|