手动AB实验显著性统计

代码如下

import numpy as np
from scipy import stats

#先将对照组与实验组的数据放入use_time_base与use_time_exp
a = []
b = []
count = 1000
for i in range(count):
    r = random.random()
    if r < 0.9:
        a.append(0)
    else:
        a.append(int((r-0.9)*100/2))
    r = random.random()
    if r < 0.89:
        b.append(0)
    else:
        b.append(int((r-0.9)*100/2))
use_time_base = np.array(a)
use_time_exp = np.array(b)


#显著度计算
print('【use_time样本概况】')
mean_base = use_time_base.mean()
std_base = use_time_base.std()
mean_exp = use_time_exp.mean()
std_exp = use_time_exp.std()
print("对照组：样本均值为%.2f，样本标准差为%.2f" % (mean_base,std_base))
print("实验组：样本均值为%.2f，样本标准差为%.2f" % (mean_exp,std_exp))
if stats.kstest(use_time_base, 'norm', (mean_base, std_base))[1] < 0.05 and stats.kstest(use_time_exp, 'norm', (mean_exp, std_exp))[1] < 0.05:
    print "对照组服从正态分布：",stats.kstest(use_time_base, 'norm', (mean_base, std_base))
    print "实验组服从正态分布：", stats.kstest(use_time_exp, 'norm', (mean_exp, std_exp))
    if stats.levene(use_time_base, use_time_exp)[1] > 0.05:
        print "对照组和实验组方差相等：", stats.levene(use_time_base, use_time_exp)
        if stats.ttest_ind(use_time_base,use_time_exp)[1] < 0.05:
            print "T检验结论：对照组和实验组存在显著差异",stats.ttest_ind(use_time_base,use_time_exp)
        else:
            print "T检验结论：对照组和实验组不存在显著差异",stats.ttest_ind(use_time_base,use_time_exp)
    else:
        print "对照组和实验组方差不相等：", stats.levene(use_time_base, use_time_exp)
        if stats.ttest_ind(use_time_base,use_time_exp,equal_var=False)[1] < 0.05:
            print "T检验结论：对照组和实验组存在显著差异",stats.ttest_ind(use_time_base,use_time_exp,equal_var=False)
        else:
            print "T检验结论：对照组和实验组不存在显著差异",stats.ttest_ind(use_time_base,use_time_exp,equal_var=False)
        if stats.mannwhitneyu(use_time_base, use_time_exp, alternative='two-sided')[1] < 0.05:
            print "非参数检验结论：对照组和实验组存在显著差异",stats.mannwhitneyu(use_time_base, use_time_exp, alternative='two-sided')
        else:
            print "非参数检验结论：对照组和实验组不存在显著差异",stats.mannwhitneyu(use_time_base, use_time_exp, alternative='two-sided')
else:
    print "样本数据不服从正态分布，进行非参数检验：",stats.mannwhitneyu(use_time_base, use_time_exp, alternative='two-sided')
    if stats.mannwhitneyu(use_time_base, use_time_exp, alternative='two-sided')[1] < 0.05:
        print("非参数检验结论：对照组和实验组存在显著差异")
    else:
        print("非参数检验结论：对照组和实验组不存在显著差异")

参考文献：

https://zhuanlan.zhihu.com/p/235826647