|
|
本帖最后由 Menuett 于 2013-12-22 15:59 编辑 & d! e- \/ ?6 s _ P
煮酒正熟 发表于 2013-12-20 12:05 ![]()
" j8 \' `8 ?8 ]' S( e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 ... # a' t! t& J& [- a4 b
L4 {+ H$ T7 c$ T( t5 X
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)。
; T) R$ y' q, B
; W. `/ u8 _! {. t' Q! a) R7 ^9 \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)
9 i2 {+ a. J; u% g- X. t9 }( n6 X+ \4 j6 S: n. C( L
R example:- q+ x! _6 _5 ]+ v) z9 g C
9 s( E; B6 S8 e8 ]
> M<-as.table(rbind(c(1668,5173),c(287,930)))
, k1 j7 T6 S7 o3 i7 T> chisq.test(M). R; C' y2 y; G6 o- X
O% K6 _( b/ K4 Q/ O! a
Pearson's Chi-squared test with Yates' continuity correction
# B7 D; c1 r {3 s2 a; g. C
# G. x% e0 T- X8 c* c1 I4 a; T) X' hdata: M
) J! {5 W8 l+ PX-squared = 0.3175, df = 1, p-value = 0.57316 ^% G# h) {( O( E3 \7 m
9 D' r" m$ m. j) J( w ]
Python example:( K& v8 q& E: w/ O1 p, o7 i
: J4 f. L. H9 f% o6 f J>>> from scipy import stats% c. q$ ~0 S( l% b1 h( P0 h3 n- C
>>> stats.chi2_contingency([[6841-5173,5173],[1217-930,930]])
2 B0 J" I9 }% M# } G T) A(0.31748297614660292, 0.57312422493552839, 1, array([[ 1659.73628692, 5181.26371308],
6 [6 W2 f7 a1 B1 w: q% Q% x [ 295.26371308, 921.73628692]])) |
|