Python: Two-Sample T-Test & Wilcoxon's Rank-Sum Test

[안재형]

from scipy import stats as st

import numpy as np

import pandas as pd

# 데이터 dental.csv의 resp만 읽어 log-변환

logy=np.log(pd.read_csv('~/data/dental.csv').resp)

death.csv

129B

# 등분산 검정

>>> st.bartlett(logy[5:10], logy[0:5])

BartlettResult(statistic=0.9715330643707207, pvalue=0.32429813448716616)

# 두 그룹의 분산이 같은 경우

>>> st.ttest_ind(logy[5:10], logy[0:5])

Ttest_indResult(statistic=-2.521742772630017, pvalue=0.035711712354790294)

# 분산이 다른 경우

>>> st.ttest_ind(logy[5:10], logy[0:5], equal_var=False)

Ttest_indResult(statistic=-2.521742772630017, pvalue=0.042454979494495546)

### Wilcoxon's Rank-Sum Test - 정규분포 근사만 가능

>>> st.ranksums(y[5:10],y[0:5])

RanksumsResult(statistic=-1.9844852778949553, pvalue=0.04720176769014221)

SAS의 npar1way에서 Kruskal-Wallis Test와 동일한 결과입니다.

R로 같은 결과를 내려면 

> wilcox.test(resp~treatment, exact=FALSE, correct=FALSE, data=dental)

Wilcoxon rank sum test data: resp by treatment W = 3, p-value = 0.0472

alternative hypothesis: true location shift is not equal to 0