|
本帖最后由 Menuett 于 2013-12-22 15:59 编辑 9 ^3 y6 j' b% m3 i
煮酒正熟 发表于 2013-12-20 12:05 1 Y+ J d$ ~5 p
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 ... - R9 D, [7 N$ o7 e! M
( g, M9 l0 ~* I. i0 t
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)。
8 W7 e5 X! I2 l( T7 z: S* j8 R; n7 `3 |
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)
3 X4 A5 }% q, j) R1 F2 G" q9 u4 \1 h; {, _! K# I) B( R
R example:; h3 E- R# y/ W/ O7 ?
8 f; A5 X* X- P) R+ W& p> M<-as.table(rbind(c(1668,5173),c(287,930)))
! K7 V7 e7 U% c- `8 _$ H7 T3 V* X> chisq.test(M)
9 v I. b+ N5 p( A: k4 |
# ?: a! n$ M" G8 r2 c Pearson's Chi-squared test with Yates' continuity correction9 U. K0 v- ~9 M; t1 L& X
0 x' M- i6 S" D# ]: i5 Y) Ydata: M8 g3 U0 n" e }$ g5 S3 X
X-squared = 0.3175, df = 1, p-value = 0.57316 V8 U9 j a1 l# S; d$ o6 o
4 Z" E1 V, E- _: B5 r/ jPython example:& _4 |# C0 _+ j: T" S8 J+ A
% |; I7 S( X- m! F>>> from scipy import stats
- w! \) A/ X, ` H) A1 x>>> stats.chi2_contingency([[6841-5173,5173],[1217-930,930]])
) M0 a3 t$ S4 n0 S% ~* w(0.31748297614660292, 0.57312422493552839, 1, array([[ 1659.73628692, 5181.26371308],
$ ]+ V' Q1 q% U- t w% P9 B [ 295.26371308, 921.73628692]])) |
|