|
本帖最后由 Menuett 于 2013-12-22 15:59 编辑 # U! R0 a( K. J: d) j
煮酒正熟 发表于 2013-12-20 12:05
1 U6 o1 S4 h# k4 p Y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 ... ) {7 O( T2 Y. K- |. M6 ?; [+ O
6 r% Q) o/ ~1 u; W( 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)。 9 d+ j6 R0 j# l8 `$ D& ~! G
$ b# w, C9 g: o& k8 `4 c4 j8 C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)
* m/ ^$ n, \+ U2 q9 ` D, o7 \& E# _0 I" r" ]% h# I+ w1 U
R example:
$ p1 B6 t7 h: {/ W. f* w) x8 p) {
- ?" c& X2 `9 u A# M> M<-as.table(rbind(c(1668,5173),c(287,930)))1 Z9 ~4 f1 t/ W% P
> chisq.test(M)
: t4 d* J" Q) `; A% k! V
9 E1 M6 N t: @& O Pearson's Chi-squared test with Yates' continuity correction# t% r9 S, c0 G' o$ S% _$ l; u: a" G
. }. O8 V& k5 {) r/ R% i; L
data: M
/ t1 a9 ~/ y0 C5 _X-squared = 0.3175, df = 1, p-value = 0.5731. s* B0 M, c; Y! P' y9 k
5 l( a2 g$ d9 m4 e7 L
Python example:5 k4 ?# w; J" G3 A% m4 s2 _7 a
- R4 K( ?& v) p; _) n, N
>>> from scipy import stats/ N7 P" n9 ?4 F3 [7 |( q
>>> stats.chi2_contingency([[6841-5173,5173],[1217-930,930]]), f2 F3 b: T4 ~9 b& ~* v3 R
(0.31748297614660292, 0.57312422493552839, 1, array([[ 1659.73628692, 5181.26371308],
2 _6 J }! ?! g: D* z [ 295.26371308, 921.73628692]])) |
|