def test_darfur_Sensemakr():
darfur_out = main.Sensemakr(model=model,
treatment='directlyharmed',
benchmark_covariates='female',
kd=[1, 2, 3])
darfur_out.summary()
darfur_out.print()
darfur_out.ovb_minimal_reporting(format='html', display=False)
darfur_out.ovb_minimal_reporting(display=False)
ovb_contour_plot(sense_obj=darfur_out)
ovb_extreme_plot(sense_obj=darfur_out)
# info
assert (darfur_out.treatment == 'directlyharmed')
assert (darfur_out.q == 1)
assert (darfur_out.alpha == 0.05)
assert (darfur_out.reduce == True)
assert (darfur_out.sensitivity_stats['dof'] == 783)
np.testing.assert_allclose(darfur_out.sensitivity_stats['r2yd_x'],
0.02187,
atol=1e-5)
np.testing.assert_allclose(darfur_out.sensitivity_stats['rv_q'],
0.13878,
atol=1e-5)
np.testing.assert_allclose(darfur_out.sensitivity_stats['rv_qa'],
0.07626,
atol=1e-5)
data = [["1x female", "2x female", "3x female"],
[0.00916428667504862, 0.0183285733500972, 0.0274928600251459],
[0.12464092303637, 0.249324064199975, 0.374050471038094],
["directlyharmed", "directlyharmed", "directlyharmed"],
[0.0752202712144491, 0.0529151723844518, 0.0303960234641548],
[0.0218733277437572, 0.0203500620779637, 0.0186700648170924],
[3.43890386024675, 2.60024623913809, 1.62806202131271],
[0.032282966, 0.012968035, -0.006253282],
[0.11815758, 0.09286231, 0.06704533]]
data = np.array(data).T.tolist()
df = pd.DataFrame(data,
columns=[
"bound_label", "r2dz_x", "r2yz_dx", "treatment",
"adjusted_estimate", "adjusted_se", "adjusted_t",
"adjusted_lower_CI", "adjusted_upper_CI"
])
df[['r2dz_x','r2yz_dx','adjusted_estimate','adjusted_se','adjusted_t','adjusted_lower_CI','adjusted_upper_CI']] =\
df[['r2dz_x','r2yz_dx','adjusted_estimate','adjusted_se','adjusted_t','adjusted_lower_CI','adjusted_upper_CI']].astype(float)
assert (darfur_out.bounds.round(6).equals(df.round(6)))
darfur_out2 = main.Sensemakr(model=model, treatment='directlyharmed')
darfur_out3 = main.Sensemakr(model=model,
treatment='directlyharmed',
q=1.0,
alpha=0.05,
reduce=True)
darfur_out3.ovb_minimal_reporting(format='html', display=False)
darfur_out3.ovb_minimal_reporting(display=False)
ovb_contour_plot(sense_obj=darfur_out2)
ovb_extreme_plot(sense_obj=darfur_out2)
def test_group_bench():
# exact
n = 1000
X = list(range(1, n + 1))
z1 = resid_maker(n, pd.DataFrame({'x': X}))
z2 = resid_maker(n, pd.DataFrame({'Z1': z1}))
x1 = resid_maker(n, pd.DataFrame({'Z1': z1, 'Z2': z2}))
x2 = resid_maker(n, pd.DataFrame({'Z1': z1, 'Z2': z2, 'X1': x1}))
d = 2 * x1 + x2 + 2 * z1 + z2 + resid_maker(
n, pd.DataFrame({
'Z1': z1,
'Z2': z2,
'X1': x1,
"X2": x2
})) * 5
y = 2 * x1 + x2 + 2 * z1 + z2 + resid_maker(
n, pd.DataFrame({
'Z1': z1,
'Z2': z2,
'X1': x1,
"X2": x2,
'D': d
})) * 5
df = pd.DataFrame({'z1': z1, 'z2': z2, 'x1': x1, 'x2': x2, 'd': d, 'y': y})
model = smf.ols(formula='y~d+x1+x2', data=df).fit()
model_dz = smf.ols(formula='y~d+z1+z2', data=df).fit()
r2yx = group_partial_r2(model=model, covariates=['x1', 'x2'])
r2yz = group_partial_r2(model=model_dz, covariates=['z1', 'z2'])
ky = r2yz / r2yx
np.testing.assert_allclose(ky, 1, atol=1e-7)
model_d = smf.ols(formula='d~x1+x2', data=df).fit()
model_dz = smf.ols(formula='d~z1+z2', data=df).fit()
r2dx = group_partial_r2(model=model_d, covariates=['x1', 'x2'])
r2dz = group_partial_r2(model=model_dz, covariates=['z1', 'z2'])
kd = r2dz / r2dx
np.testing.assert_allclose(kd, 1, atol=1e-7)
out = main.Sensemakr(model=model,
treatment="d",
benchmark_covariates=[['x1', 'x2']],
kd=kd,
ky=ky)
out2 = main.Sensemakr(
model=model,
treatment="d",
benchmark_covariates={'[' + "'x1', 'x2'" + ']': ['x1', 'x2']},
kd=kd,
ky=ky)
out3 = main.Sensemakr(
model=model,
treatment="d",
benchmark_covariates={'[' + "'x1', 'x2'" + ']': ['x1', 'x2']},
kd=[1, 2, 3])
bound = ovb_partial_r2_bound(model=model, treatment="d")
bound2 = ovb_partial_r2_bound(model=model,
treatment="d",
benchmark_covariates='x1')
assert (out.bounds.equals(out2.bounds))
def test_darfur_manual_bounds():
sense_out = main.Sensemakr(model=model,
treatment='directlyharmed',
benchmark_covariates='female',
r2dz_x=0.1)
sense_out.summary()
bounds_check = sense_out.bounds
to_check = bounds_check.adjusted_se
true_check = adjusted_se(model=model,
treatment='directlyharmed',
r2dz_x=0.1,
r2yz_dx=0.1)
assert (to_check.values[0] == true_check)
def test_darfur_different_q():
darfur_out = main.Sensemakr(model=model,
treatment='directlyharmed',
benchmark_covariates='female',
q=2,
kd=[1, 2, 3])
rvq = darfur_out.sensitivity_stats['rv_q']
rvqa = darfur_out.sensitivity_stats['rv_qa']
assert (rvq == robustness_value(model=model,
covariates='directlyharmed',
q=2).values)
assert (rvqa == robustness_value(model=model,
covariates='directlyharmed',
q=2,
alpha=0.05).values)
def test_darfur_sensemakr_manually():
model_treat = smf.ols(
formula='directlyharmed ~ age + farmer_dar + herder_dar +\
pastvoted + hhsize_darfur + female + village',
data=darfur).fit()
darfur_out = main.Sensemakr(estimate=0.09731582,
se=0.02325654,
dof=783,
treatment="directlyharmed",
benchmark_covariates="female",
r2dxj_x=partial_r2(model_treat,
covariates="female"),
r2yxj_dx=partial_r2(model,
covariates="female"),
kd=[1, 2, 3])
assert (darfur_out.q == 1)
assert (darfur_out.alpha == 0.05)
assert (darfur_out.treatment == "D")
assert (darfur_out.reduce == True)
assert (darfur_out.sensitivity_stats['dof'] == 783)
np.testing.assert_allclose(darfur_out.sensitivity_stats['r2yd_x'],
0.02187,
atol=1e-4)
np.testing.assert_allclose(darfur_out.sensitivity_stats['rv_q'],
0.13878,
atol=1e-5)
np.testing.assert_allclose(darfur_out.sensitivity_stats['rv_qa'],
0.07626,
atol=1e-5)
data = [["1x female", "2x female", "3x female"],
[0.00916428667504862, 0.0183285733500972, 0.0274928600251459],
[0.12464092303637, 0.249324064199975, 0.374050471038094],
[0.0752202698486415, 0.0529151689180575, 0.0303960178770157],
[0.0218733298036818, 0.0203500639944344, 0.0186700665753491],
[3.43890347394571, 2.60024582392121, 1.62806156873318],
[0.0322829603180086, 0.0129680276030601, -0.00625329133645187],
[0.118157579379274, 0.092862310233055, 0.0670453270904833]]
data = np.array(data).T.tolist()
df = pd.DataFrame(data,
columns=[
"bound_label", "r2dz_x", "r2yz_dx",
"adjusted_estimate", "adjusted_se", "adjusted_t",
"adjusted_lower_CI", "adjusted_upper_CI"
])
df[['r2dz_x','r2yz_dx','adjusted_estimate','adjusted_se','adjusted_t','adjusted_lower_CI','adjusted_upper_CI']] =\
df[['r2dz_x','r2yz_dx','adjusted_estimate','adjusted_se','adjusted_t','adjusted_lower_CI','adjusted_upper_CI']].astype(float)
assert (darfur_out.bounds.round(6).equals(df.round(6)))
def test_manual_input():
out = main.Sensemakr(estimate=2,
se=2,
dof=100,
r2dz_x=0.2,
r2dxj_x=0.2,
r2yxj_dx=0.2,
kd=2,
ky=1)
out.summary()
est = out.bounds['adjusted_estimate'][0]
est_check = adjusted_estimate(estimate=2,
se=2,
dof=100,
r2dz_x=0.2,
r2yz_dx=0.2)
assert (est == est_check)
def test_darfur_group_benchmarks():
village = [k for k in model.params.index.values if 'village' in k]
sensitivity = main.Sensemakr(model=model,
treatment='directlyharmed',
benchmark_covariates=[village],
kd=0.3)
r2y = group_partial_r2(model, covariates=village)
treat_model = smf.ols(
formula='directlyharmed ~ age + farmer_dar + herder_dar +\
pastvoted + hhsize_darfur + female + village',
data=darfur).fit()
r2d = group_partial_r2(treat_model, covariates=village)
bounds_check = ovb_partial_r2_bound(r2dxj_x=r2d,
r2yxj_dx=r2y,
kd=0.3,
benchmark_covariates=['manual'])
bounds = sensitivity.bounds
np.testing.assert_allclose(bounds_check['r2dz_x'].values,
bounds['r2dz_x'].values,
atol=1e-5)
np.testing.assert_allclose(bounds_check['r2yz_dx'].values,
bounds['r2yz_dx'].values,
atol=1e-5)
model = smf.ols(
formula='peacefactor ~ directlyharmed + age + farmer_dar + herder_dar +\
pastvoted + hhsize_darfur + female + village',
data=darfur).fit()
treatment = "directlyharmed"
q = 1.0
alpha = 0.05
reduce = True
benchmark_covariates = ["female"]
kd = [1, 2, 3]
ky = kd
s = main.Sensemakr(model,
treatment,
q=q,
alpha=alpha,
reduce=reduce,
benchmark_covariates=benchmark_covariates,
kd=kd)
s2 = main.Sensemakr(model,
treatment,
q=q,
alpha=alpha,
reduce=False,
benchmark_covariates=benchmark_covariates,
kd=kd)
def test_plots():
ovb_contour_plot(model=model, treatment='directlyharmed', r2dz_x=0.1)
ovb_contour_plot(model=model, treatment='directlyharmed')