def test_epsilon(self):
"""Test function epsilon."""
df_pivot = df.pivot(index='Subject', columns='Time',
values='Scores').reset_index(drop=True)
eps_gg = epsilon(df_pivot)
eps_hf = epsilon(df_pivot, correction='hf')
eps_lb = epsilon(df_pivot, correction='lb')
# Compare with ezANOVA
assert np.allclose([eps_gg, eps_hf, eps_lb], [0.9987509, 1, 0.5])
def test_epsilon(self):
"""Test function epsilon."""
df_pivot = df.pivot(index='Subject', columns='Time',
values='Scores').reset_index(drop=True)
eps_gg = epsilon(df_pivot)
eps_hf = epsilon(df_pivot, correction='hf')
eps_lb = epsilon(df_pivot, correction='lb')
# In long-format
eps_gg_rm = epsilon(df, subject='Subject', within='Time', dv='Scores')
eps_hf_rm = epsilon(df,
subject='Subject',
within='Time',
dv='Scores',
correction='hf')
assert np.isclose(eps_gg_rm, eps_gg)
assert np.isclose(eps_hf_rm, eps_hf)
# Compare with ezANOVA
assert np.allclose([eps_gg, eps_hf, eps_lb], [0.9987509, 1, 0.5])
# Time has only two values so epsilon is one.
assert epsilon(pa, correction='lb') == epsilon(pa, correction='gg')
assert epsilon(pa, correction='gg') == epsilon(pa, correction='hf')
# Lower bound <= Greenhouse-Geisser <= Huynh-Feldt
assert epsilon(pb, correction='lb') <= epsilon(pb, correction='gg')
assert epsilon(pb, correction='gg') <= epsilon(pb, correction='hf')
assert epsilon(pab, correction='lb') <= epsilon(pab, correction='gg')
assert epsilon(pab, correction='gg') <= epsilon(pab, correction='hf')
# Lower bound == 0.5 for pb and pab
assert epsilon(pb, correction='lb') == epsilon(pab, correction='lb')
assert np.allclose(epsilon(pb), 0.9691030) # ez
assert np.allclose(epsilon(pb, correction='hf'), 1.0) # ez
# Epsilon for the interaction (shape = (2, N))
assert np.allclose(epsilon(pab), 0.7271664)
assert np.allclose(epsilon(pab, correction='hf'), 0.831161)
assert epsilon(pab) == epsilon(pab.swaplevel(axis=1))
assert epsilon(pab_single) == epsilon(pab_single.swaplevel(axis=1))
eps_gg_rm = epsilon(data,
subject='Subject',
dv='Performance',
within=['Time', 'Metric'])
assert eps_gg_rm == epsilon(pab)
# Now with a (3, 4) two-way design
assert np.allclose(epsilon(pa1), 0.9963275)
assert np.allclose(epsilon(pa1, correction='hf'), 1.)
assert np.allclose(epsilon(pb1), 0.9716288)
assert np.allclose(epsilon(pb1, correction='hf'), 1.)
assert 0.8 < epsilon(pab1) < .90 # Pingouin = .822, ez = .856
eps_gg_rm = epsilon(df3,
subject='subj',
dv='dv',
within=['within1', 'within2'])
assert eps_gg_rm == epsilon(pab1)
# With missing values
eps_gg_rm = epsilon(df_nan,
subject='Subject',
within='Time',
dv='Scores')
# 3 repeated measures factor
with pytest.raises(ValueError):
epsilon(pab_3fac)