|
本帖最后由 Menuett 于 2013-12-22 15:59 编辑 $ l2 I( u2 Q& G: w+ H1 y6 N# V
煮酒正熟 发表于 2013-12-20 12:05 9 B+ }, h, l# P# G: V
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 ...
5 |9 Y, n$ ?( r4 Q
; m) v* X* M" d5 A# d5 p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)。
$ h3 O7 z) D+ W/ y i) ?: G% F- Y( U* T7 v$ E) f9 P! A) l
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), k. u. P- z) U: E$ K& \
4 A# ^! ]2 `9 ?5 _# u" M3 t- z
R example:! ^9 B$ @' h' i i( u/ \$ e
( I, G$ P! w' E! F# }> M<-as.table(rbind(c(1668,5173),c(287,930)))
' U6 J# g( r6 u# K/ G> chisq.test(M)
1 `; l4 Y: b8 v' Q7 f* B' K
* E. {$ C' f7 M! I Pearson's Chi-squared test with Yates' continuity correction
% Z3 X3 V9 F. X+ M" C3 M7 }
# Z3 q' }% D% Ydata: M. v$ c& H% i+ y( N
X-squared = 0.3175, df = 1, p-value = 0.5731
% y( O8 z& \2 {8 q* N [
: X; ^' C! o1 _Python example:
1 I0 H$ | Q# Q8 |' R! N" Y1 \5 M$ o4 w' D9 N* D: w7 F
>>> from scipy import stats) u3 @ U9 s/ R7 e6 e4 \
>>> stats.chi2_contingency([[6841-5173,5173],[1217-930,930]])# g8 }" N* T% T$ w% }
(0.31748297614660292, 0.57312422493552839, 1, array([[ 1659.73628692, 5181.26371308],5 w, q. G& g- N: d# @ r u. q& O
[ 295.26371308, 921.73628692]])) |
|