def test_mediation_analysis(self):
"""Test function mediation_analysis.
"""
ma = mediation_analysis(data=df, x='X', m='M', y='Y', n_boot=500)
# Compare against R package mediation
assert_equal(ma['coef'].values,
[0.5610, 0.6542, 0.3961, 0.0396, 0.3565])
_, dist = mediation_analysis(data=df,
x='X',
m='M',
y='Y',
n_boot=1000,
return_dist=True)
assert dist.size == 1000
mediation_analysis(data=df, x='X', m='M', y='Y', alpha=0.01)
# Check with a binary mediator
ma = mediation_analysis(data=df, x='X', m='Mbin', y='Y', n_boot=2000)
assert_almost_equal(ma['coef'][0], -0.0208, decimal=2)
# Indirect effect
assert_almost_equal(ma['coef'][4], 0.0033, decimal=2)
assert ma['sig'][4] == 'No'
# Direct effect
assert_almost_equal(ma['coef'][3], 0.3956, decimal=2)
assert_almost_equal(ma['CI[2.5%]'][3], 0.1714, decimal=2)
assert_almost_equal(ma['CI[97.5%]'][3], 0.617, decimal=1)
assert ma['sig'][3] == 'Yes'
# With multiple mediator
np.random.seed(42)
df.rename(columns={"M": "M1"}, inplace=True)
df['M2'] = np.random.randint(0, 10, df.shape[0])
ma2 = mediation_analysis(data=df,
x='X',
m=['M1', 'M2'],
y='Y',
seed=42)
assert ma['coef'][2] == ma2['coef'][4]
# With covariate
mediation_analysis(data=df, x='X', m='M1', y='Y', covar='M2')
mediation_analysis(data=df, x='X', m='M1', y='Y', covar=['M2'])
mediation_analysis(data=df,
x='X',
m=['M1', 'Ybin'],
y='Y',
covar=['Mbin', 'M2'])
# Test helper function _pval_from_bootci
np.random.seed(123)
bt2 = np.random.normal(loc=2, size=1000)
bt3 = np.random.normal(loc=3, size=1000)
assert _pval_from_bootci(bt2, 0) == 1
assert _pval_from_bootci(bt2, 0.9) < 0.10
assert _pval_from_bootci(bt3, 0.9) < _pval_from_bootci(bt2, 0.9)
def test_mediation_analysis(self):
"""Test function mediation_analysis.
"""
ma = mediation_analysis(data=df, x='X', m='M', y='Y', n_boot=500)
# Compare against R package mediation
assert_equal(ma['coef'].round(4).to_numpy(),
[0.5610, 0.6542, 0.3961, 0.0396, 0.3565])
_, dist = mediation_analysis(data=df,
x='X',
m='M',
y='Y',
n_boot=1000,
return_dist=True)
assert dist.size == 1000
mediation_analysis(data=df, x='X', m='M', y='Y', alpha=0.01)
# Check with a binary mediator
ma = mediation_analysis(data=df, x='X', m='Mbin', y='Y', n_boot=2000)
assert_almost_equal(ma['coef'][0], -0.0208, decimal=2)
# Indirect effect
assert_almost_equal(ma['coef'][4], 0.0033, decimal=2)
assert ma['sig'][4] == 'No'
# Direct effect
assert_almost_equal(ma['coef'][3], 0.3956, decimal=2)
assert_almost_equal(ma['CI[2.5%]'][3], 0.1714, decimal=2)
assert_almost_equal(ma['CI[97.5%]'][3], 0.617, decimal=1)
assert ma['sig'][3] == 'Yes'
# Check if `logreg_kwargs` is being passed on to `LogisticRegression`
with pytest.raises(ValueError):
mediation_analysis(data=df,
x='X',
m='Mbin',
y='Y',
n_boot=2000,
logreg_kwargs=dict(max_iter=-1))
# Solve with 0 iterations and make sure that the results are different
ma = mediation_analysis(data=df,
x='X',
m='Mbin',
y='Y',
n_boot=2000,
logreg_kwargs=dict(max_iter=0))
with pytest.raises(AssertionError):
assert_almost_equal(ma['coef'][0], -0.0208, decimal=2)
with pytest.raises(AssertionError):
assert_almost_equal(ma['coef'][4], 0.0033, decimal=3)
# With multiple mediator
np.random.seed(42)
df.rename(columns={"M": "M1"}, inplace=True)
df['M2'] = np.random.randint(0, 10, df.shape[0])
ma2 = mediation_analysis(data=df,
x='X',
m=['M1', 'M2'],
y='Y',
seed=42)
assert ma['coef'][2] == ma2['coef'][4]
# With covariate
mediation_analysis(data=df, x='X', m='M1', y='Y', covar='M2')
mediation_analysis(data=df, x='X', m='M1', y='Y', covar=['M2'])
mediation_analysis(data=df,
x='X',
m=['M1', 'Ybin'],
y='Y',
covar=['Mbin', 'M2'])
# Test helper function _pval_from_bootci
np.random.seed(123)
bt2 = np.random.normal(loc=2, size=1000)
bt3 = np.random.normal(loc=3, size=1000)
assert _pval_from_bootci(bt2, 0) == 1
assert _pval_from_bootci(bt2, 0.9) < 0.10
assert _pval_from_bootci(bt3, 0.9) < _pval_from_bootci(bt2, 0.9)