def test_levene(self):
data = self.data
# lawstat: Test Statistic = 1.0866123063642, p-value = 0.3471072204516
statistic = 1.0866123063642
p_value = 0.3471072204516
res0 = smo.test_scale_oneway(data,
method='equal',
center='median',
transform='abs',
trim_frac_mean=0.2)
assert_allclose(res0.pvalue, p_value, rtol=1e-13)
assert_allclose(res0.statistic, statistic, rtol=1e-13)
# library car
# > lt = leveneTest(y ~ g, df3, center=mean, trim=0.2)
statistic = 1.10732113109744
p_value = 0.340359251994645
df = [2, 40]
res0 = smo.test_scale_oneway(data,
method='equal',
center='trimmed',
transform='abs',
trim_frac_mean=0.2)
assert_allclose(res0.pvalue, p_value, rtol=1e-13)
assert_allclose(res0.statistic, statistic, rtol=1e-13)
assert_allclose(res0.df, df)
# library(onewaytests)
# test uses mean as center
# > st = homog.test(y ~ g, df3)
statistic = 1.07894485177512
parameter = [2, 40] # df
p_value = 0.349641166869223
# method = "Levene's Homogeneity Test"
res0 = smo.test_scale_oneway(data,
method='equal',
center='mean',
transform='abs',
trim_frac_mean=0.2)
assert_allclose(res0.pvalue, p_value, rtol=1e-13)
assert_allclose(res0.statistic, statistic, rtol=1e-13)
assert_allclose(res0.df, parameter)
# > st = homog.test(y ~ g, df3, method = "Bartlett")
statistic = 3.01982414477323
# parameter = 2 # scipy bartlett does not return df
p_value = 0.220929402900495
# method = "Bartlett's Homogeneity Test"
# Bartlett is in scipy.stats
from scipy import stats
stat, pv = stats.bartlett(*data)
assert_allclose(pv, p_value, rtol=1e-13)
assert_allclose(stat, statistic, rtol=1e-13)
mod = _MultivariateOLS.from_formula('人均地区生产总值+公共财政支出 ~ 省份', data)
result = mod.fit(method='svd')
result.mv_test()
#3.多元统计-协方差阵检验
temp_data = []
temp_name = []
for name, group in data[['省份', '人均地区生产总值', '公共财政支出']].groupby(['省份']):
temp_data.append(np.cov(np.asarray(group[['人均地区生产总值', '公共财政支出']].T)))
temp_name.append(name)
#statistic_base 是Box's M统计量
#pvalue是书中给出的p值
test_cov_oneway(temp_data, [5, 5, 5])
#4.误差方差分析
temp_data = []
temp_name = []
for name, group in data[['省份', '人均地区生产总值', '公共财政支出']].groupby(['省份']):
temp_data.append(np.array((group['人均地区生产总值'] - group['人均地区生产总值'].mean())))
# temp_name.append(name)
temp_data = np.array(temp_data)
res0 = smo.test_scale_oneway(temp_data,
method='equal',
center='mean',
transform='abs',
trim_frac_mean=0.2)
'statistic:%s,df2:%s,Sig.:%s,pvalue:%s' % (res0.statistic, res0.df_num,
res0.df_denom, res0.pvalue)
def test_options(self):
# regression tests for options,
# many might not be implemented in other packages
data = self.data
# regression numbers from initial run
statistic, p_value = 1.0173464626246675, 0.3763806150460239
df = (2.0, 24.40374758005409)
res = smo.test_scale_oneway(data,
method='unequal',
center='median',
transform='abs',
trim_frac_mean=0.2)
assert_allclose(res.pvalue, p_value, rtol=1e-13)
assert_allclose(res.statistic, statistic, rtol=1e-13)
assert_allclose(res.df, df)
statistic, p_value = 1.0329722145270606, 0.3622778213868562
df = (1.83153791573948, 30.6733640949525)
p_value2 = 0.3679999679787619
df2 = (2, 30.6733640949525)
res = smo.test_scale_oneway(data,
method='bf',
center='median',
transform='abs',
trim_frac_mean=0.2)
assert_allclose(res.pvalue, p_value, rtol=1e-13)
assert_allclose(res.statistic, statistic, rtol=1e-13)
assert_allclose(res.df, df)
assert_allclose(res.pvalue2, p_value2, rtol=1e-13)
assert_allclose(res.df2, df2)
statistic, p_value = 1.7252431333701745, 0.19112038168209514
df = (2.0, 40.0)
res = smo.test_scale_oneway(data,
method='equal',
center='mean',
transform='square',
trim_frac_mean=0.2)
assert_allclose(res.pvalue, p_value, rtol=1e-13)
assert_allclose(res.statistic, statistic, rtol=1e-13)
assert_equal(res.df, df)
statistic, p_value = 0.4129696057329463, 0.6644711582864451
df = (2.0, 40.0)
res = smo.test_scale_oneway(
data,
method='equal',
center='mean',
transform=lambda x: np.log(x * x), # noqa
trim_frac_mean=0.2)
assert_allclose(res.pvalue, p_value, rtol=1e-13)
assert_allclose(res.statistic, statistic, rtol=1e-13)
assert_allclose(res.df, df)
# compare no transform with standard anova
res = smo.test_scale_oneway(data,
method='unequal',
center=0,
transform='identity',
trim_frac_mean=0.2)
res2 = anova_oneway(self.data, use_var="unequal")
assert_allclose(res.pvalue, res2.pvalue, rtol=1e-13)
assert_allclose(res.statistic, res2.statistic, rtol=1e-13)
assert_allclose(res.df, res2.df)