AIM-PIbd-32-Kurbanova-A-A/aimenv/Lib/site-packages/statsmodels/treatment/tests/test_teffects.py
2024-10-02 22:15:59 +04:00

125 lines
4.6 KiB
Python

"""
Created on Feb 3, 2022 1:04:22 PM
Author: Josef Perktold
License: BSD-3
"""
import os
import numpy as np
from numpy.testing import assert_allclose
import pandas as pd
import pytest
from statsmodels.regression.linear_model import OLS
from statsmodels.discrete.discrete_model import Probit
from statsmodels.treatment.treatment_effects import (
TreatmentEffect
)
from .results import results_teffects as res_st
cur_dir = os.path.abspath(os.path.dirname(__file__))
file_name = 'cataneo2.csv'
file_path = os.path.join(cur_dir, 'results', file_name)
dta_cat = pd.read_csv(file_path)
formula = 'mbsmoke_ ~ mmarried_ + mage + mage2 + fbaby_ + medu'
res_probit = Probit.from_formula(formula, dta_cat).fit()
methods = [
("ra", res_st.results_ra),
("ipw", res_st.results_ipw),
("aipw", res_st.results_aipw),
("aipw_wls", res_st.results_aipw_wls),
("ipw_ra", res_st.results_ipwra),
]
class TestTEffects():
@classmethod
def setup_class(cls):
formula_outcome = 'bweight ~ prenatal1_ + mmarried_ + mage + fbaby_'
mod = OLS.from_formula(formula_outcome, dta_cat)
tind = np.asarray(dta_cat['mbsmoke_'])
cls.teff = TreatmentEffect(mod, tind, results_select=res_probit)
def test_aux(self):
prob = res_probit.predict()
assert prob.shape == (4642,)
@pytest.mark.parametrize('case', methods)
def test_effects(self, case):
meth, res2 = case
teff = self.teff
res1 = getattr(teff, meth)(return_results=False)
assert_allclose(res1[:2], res2.table[:2, 0], rtol=1e-4)
# if meth in ["ipw", "aipw", "aipw_wls", "ra", "ipw_ra"]:
res0 = getattr(teff, meth)(return_results=True)
assert_allclose(res1, res0.effect, rtol=1e-4)
res1 = res0.results_gmm
# TODO: check ra and ipw difference 5e-6, others pass at 1e-12
assert_allclose(res0.start_params, res1.params, rtol=1e-5)
assert_allclose(res1.params[:2], res2.table[:2, 0], rtol=1e-5)
assert_allclose(res1.bse[:2], res2.table[:2, 1], rtol=1e-3)
assert_allclose(res1.tvalues[:2], res2.table[:2, 2], rtol=1e-3)
assert_allclose(res1.pvalues[:2], res2.table[:2, 3],
rtol=1e-4, atol=1e-15)
ci = res1.conf_int()
assert_allclose(ci[:2, 0], res2.table[:2, 4], rtol=5e-4)
assert_allclose(ci[:2, 1], res2.table[:2, 5], rtol=5e-4)
# test all GMM params
# constant is in different position in Stata, `idx` rearanges
k_p = len(res1.params)
if k_p == 8:
# IPW, no outcome regression
idx = [0, 1, 7, 2, 3, 4, 5, 6]
elif k_p == 18:
idx = [0, 1, 6, 2, 3, 4, 5, 11, 7, 8, 9, 10, 17, 12, 13, 14,
15, 16]
elif k_p == 12:
# RA, no selection regression
idx = [0, 1, 6, 2, 3, 4, 5, 11, 7, 8, 9, 10]
else:
idx = np.arange(k_p)
# TODO: check if improved optimization brings values closer
assert_allclose(res1.params, res2.table[idx, 0], rtol=1e-4)
assert_allclose(res1.bse, res2.table[idx, 1], rtol=0.05)
# test effects on the treated, not available for aipw
if not meth.startswith("aipw"):
table = res2.table_t
res1 = getattr(teff, meth)(return_results=False, effect_group=1)
assert_allclose(res1[:2], table[:2, 0], rtol=1e-4)
res0 = getattr(teff, meth)(return_results=True, effect_group=1)
# TODO: check ipw difference 1e-5, others pass at 1e-12
assert_allclose(res1, res0.effect, rtol=2e-5)
res1 = res0.results_gmm
# TODO: check ra difference 4e-5, others pass at 1e-12
assert_allclose(res0.start_params, res1.params, rtol=5e-5)
assert_allclose(res1.params[:2], table[:2, 0], rtol=5e-5)
assert_allclose(res1.bse[:2], table[:2, 1], rtol=1e-3)
assert_allclose(res1.tvalues[:2], table[:2, 2], rtol=1e-3)
assert_allclose(res1.pvalues[:2], table[:2, 3],
rtol=1e-4, atol=1e-15)
ci = res1.conf_int()
assert_allclose(ci[:2, 0], table[:2, 4], rtol=5e-4)
assert_allclose(ci[:2, 1], table[:2, 5], rtol=5e-4)
# consistency check, effect on untreated, not in Stata
res1 = getattr(teff, meth)(return_results=False, effect_group=0)
res0 = getattr(teff, meth)(return_results=True, effect_group=0)
assert_allclose(res1, res0.effect, rtol=1e-12)
assert_allclose(res0.start_params, res0.results_gmm.params,
rtol=1e-12)