|
本帖最后由 Menuett 于 2013-12-22 15:59 编辑 6 J9 q9 F- q0 D- v1 U, {9 `* \6 \
煮酒正熟 发表于 2013-12-20 12:05 ![]()
. y. P I* Q% m9 u" X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 ... ! n$ N3 b. V1 N% E- W; {- X1 D* G# M
& L& s8 ]: C3 B+ O) {" M! ]6 K
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)。 5 S6 P6 |( A) V0 D6 }* t
; L5 C! r/ Y" {/ N
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)+ J# } R! t6 v C, W! b
' F1 F! ], b5 g2 D: H4 Z) u) O4 K, I/ AR example:
# Q+ I1 E8 l0 Z) C1 d* e
5 ^8 H5 Z$ a0 t4 ?" K! ^$ I> M<-as.table(rbind(c(1668,5173),c(287,930)))
9 Z7 Q( k5 Q! q" g# J3 e+ \2 s! u> chisq.test(M)! o F* u% K$ W& y
' N* N, V- P% T* h/ @ Pearson's Chi-squared test with Yates' continuity correction9 z* p& f8 s1 r9 E
2 a$ b. q4 R3 K E: v/ ~
data: M
: w8 f6 H; f. H. |" ZX-squared = 0.3175, df = 1, p-value = 0.5731" S( p3 D" W9 K& C1 U5 K$ u. p
! m8 A# n/ c$ \- ]! p/ L9 ]
Python example:
( g" M" F( _/ ]2 Q. k+ b6 L, C
! a3 C M: h4 Z+ ]>>> from scipy import stats
; x+ z0 \" e, X>>> stats.chi2_contingency([[6841-5173,5173],[1217-930,930]])# I% ^0 w# }# ^9 e8 m! y- S
(0.31748297614660292, 0.57312422493552839, 1, array([[ 1659.73628692, 5181.26371308]," u* `1 e$ o& L* e0 S
[ 295.26371308, 921.73628692]])) |
|