|
本帖最后由 Menuett 于 2013-12-22 15:59 编辑 1 l- J- a& G" x
煮酒正熟 发表于 2013-12-20 12:05
" p: I9 w: y( k6 |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 ... # W$ v/ z! g/ R3 H: E! R) T3 ?# C6 @
! w/ ^% T& w- n: a G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)。 $ ], v8 M3 a7 o" v3 F# G) r
* Y u' e9 Y* i+ J5 Z @/ E- S% 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)2 ^' ?' d& x8 s; H2 i7 T
0 {% s) s# y) X8 P9 p
R example:% J$ [" }6 T o6 u# Y
6 W+ j( [5 l% N4 Y/ Z
> M<-as.table(rbind(c(1668,5173),c(287,930)))* X! f7 V6 v4 K- _
> chisq.test(M)- W: S9 I7 b7 i7 }8 r& {' z
( \( v5 m% L }, i2 e* s" W
Pearson's Chi-squared test with Yates' continuity correction
. e, d6 P% n; M' K0 s- T W6 V2 s7 B
data: M
6 R/ p9 n8 o) P! S0 I8 a) g# @. ^X-squared = 0.3175, df = 1, p-value = 0.57317 N* A S" F' T* X6 D" K) x$ Y
3 \8 \! c! H% C1 C( dPython example:
: W6 u# a; s8 a2 d
; Y0 T. A. ]+ j2 j' o>>> from scipy import stats
]+ p+ H9 m. r! \- R>>> stats.chi2_contingency([[6841-5173,5173],[1217-930,930]]), x, Q! v( B" {
(0.31748297614660292, 0.57312422493552839, 1, array([[ 1659.73628692, 5181.26371308]," S5 X4 b2 ]+ @% I% }7 j
[ 295.26371308, 921.73628692]])) |
|