def test_pairwise_gameshowell(self):
"""Test function pairwise_gameshowell.
The p-values are slightly different because of a different algorithm
used to calculate the studentized range approximation, but
significance should be the same.
"""
# Compare with R package `userfriendlyscience` - Hair color dataset
# Update Feb 2021: The userfriendlyscience package has been removed
# from CRAN.
df = read_dataset('anova')
stats = pairwise_gameshowell(dv='Pain threshold', between='Hair color',
data=df)
assert np.array_equal(np.abs(stats['T'].round(2)),
[2.47, 1.42, 1.75, 4.09, 1.11, 3.56])
assert np.array_equal(stats['df'].round(2),
[7.91, 7.94, 6.56, 8.0, 6.82, 6.77])
# JASP: [0.1401, 0.5228, 0.3715, 0.0148, 0.6980, 0.0378]
# Pingouin: [0.1401, 0.5220, 0.3722, 0.0148, 0.6848, 0.0378]
assert np.allclose([0.1401, 0.5228, 0.3715, 0.0148, 0.6980, 0.0378],
stats.loc[:, 'pval'].to_numpy().round(3),
atol=0.05)
sig = stats['pval'].apply(lambda x: 'Yes' if x < 0.05 else
'No').to_numpy()
assert np.array_equal(sig, ['No', 'No', 'No', 'Yes', 'No', 'Yes'])
# Compare with JASP in the Palmer Penguins dataset
df = read_dataset("penguins")
stats = pairwise_gameshowell(data=df, dv="body_mass_g",
between="species").round(4)
assert np.array_equal(stats['A'], ["Adelie", "Adelie", "Chinstrap"])
assert np.array_equal(stats['B'], ["Chinstrap", "Gentoo", "Gentoo"])
assert np.array_equal(stats['diff'], [-32.426, -1375.354, -1342.928])
assert np.array_equal(stats['se'], [59.7064, 58.8109, 65.1028])
assert np.array_equal(stats['df'], [152.4548, 249.6426, 170.4044])
assert np.array_equal(stats['T'], [-0.5431, -23.3860, -20.6278])
# P-values JASP: [0.8502, 0.0000, 0.0000]
# P-values Pingouin: [0.8339, 0.0010, 0.0010]
sig = stats['pval'].apply(lambda x: 'Yes' if x < 0.05 else
'No').to_numpy()
assert np.array_equal(sig, ['No', 'Yes', 'Yes'])
# Same but with balanced group
df_balanced = df.groupby('species').head(20).copy()
# To complicate things, let's encode between as a categorical
df_balanced['species'] = df_balanced['species'].astype('category')
stats = pairwise_gameshowell(data=df_balanced, dv="body_mass_g",
between="species").round(4)
assert np.array_equal(stats['A'], ["Adelie", "Adelie", "Chinstrap"])
assert np.array_equal(stats['B'], ["Chinstrap", "Gentoo", "Gentoo"])
assert np.array_equal(stats['diff'], [-142.5, -1457.5, -1315.])
assert np.array_equal(stats['se'], [104.5589, 163.1546, 154.1104])
assert np.array_equal(stats['df'], [35.5510, 30.8479, 26.4576])
assert np.array_equal(stats['T'], [-1.3629, -8.9332, -8.5328])
# P-values JASP: [0.3709, 0.0000, 0.0000]
# P-values Pingouin: [0.3719, 0.0010, 0.0010]
sig = stats['pval'].apply(lambda x: 'Yes' if x < 0.05 else
'No').to_numpy()
assert np.array_equal(sig, ['No', 'Yes', 'Yes'])
def test_pairwise_gameshowell(self):
"""Test function pairwise_gameshowell"""
df = read_dataset('anova')
stats = pairwise_gameshowell(dv='Pain threshold', between='Hair color',
data=df)
# Compare with R package `userfriendlyscience`
np.testing.assert_array_equal(np.abs(stats['T'].round(2)),
[2.48, 1.42, 1.75, 4.09, 1.11, 3.56])
np.testing.assert_array_equal(stats['df'].round(2),
[7.91, 7.94, 6.56, 8.0, 6.82, 6.77])
sig = stats['pval'].apply(lambda x: 'Yes' if x < 0.05 else 'No').values
np.testing.assert_array_equal(sig, ['No', 'No', 'No', 'Yes', 'No',
'Yes'])