|
本帖最后由 Menuett 于 2013-12-22 15:59 编辑 ! d& c4 E2 x# }' V
煮酒正熟 发表于 2013-12-20 12:05 & C4 a8 U9 a; u# g% T- Z2 a1 ^
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 ... 3 t: Y0 I; w& `! O$ j
4 ]4 n6 U. Q/ y7 K7 C
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 e0 f$ o4 L: x
) P, d: ~$ j2 o5 x' 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) B& F6 Z# E" N$ S8 c8 [' J% t0 z
) F4 A6 L7 ?, n) ?4 i) f! N$ I
R example:2 U E- n6 J M( ^% O ~1 C" [
& J6 l0 X. G% ^5 y# P> M<-as.table(rbind(c(1668,5173),c(287,930)))
9 j: l H2 _ V4 o8 S, J> chisq.test(M)
" A1 e: b, g1 k/ y" f/ Z8 S$ e- r; g7 |, d
Pearson's Chi-squared test with Yates' continuity correction
. h- g4 B; @* @# a7 i3 ?3 @4 g' s E8 V" T4 d
data: M
6 n+ R" H" ? C- y6 Z- @/ mX-squared = 0.3175, df = 1, p-value = 0.5731
2 X1 j1 N9 o% }% M' N! b/ d
. ^% H+ I2 E) \1 d2 D7 oPython example:! k4 `8 m8 H: G
" z* n" G" G3 d6 }# m8 o) E* d
>>> from scipy import stats% ^# U, f1 r* j5 Z% S- s+ P
>>> stats.chi2_contingency([[6841-5173,5173],[1217-930,930]])
6 h5 W0 q; x* N- V2 x! ?(0.31748297614660292, 0.57312422493552839, 1, array([[ 1659.73628692, 5181.26371308],
% n& [) f; ^7 L* s* C [ 295.26371308, 921.73628692]])) |
|