import pandas as pd
from statsmodels.formula.api import ols
from linearmodels.iv import IV2SLS, compare第11章のPythonコード
第11章 操作変数法
パッケージの読み込み
11.2 操作変数法の実施：2段階最小二乗法
11.2.1 2段階最小二乗法
cigdata = pd.read_csv('cigarette.csv')
result_ols = ols('bw ~ cigpacks', data = cigdata).fit()
print(result_ols.summary().tables[1])==============================================================================
coef std err t P>|t| [0.025 0.975]
------------------------------------------------------------------------------
Intercept 3423.2281 17.023 201.094 0.000 3389.823 3456.633
cigpacks -62.4019 1.544 -40.405 0.000 -65.433 -59.371
==============================================================================first_stage = ols('cigpacks ~ price', data = cigdata).fit()
cigdata = cigdata.assign(fitted = first_stage.predict())
result_iv = ols('bw ~ fitted', data = cigdata).fit()
print(result_iv.summary().tables[1])==============================================================================
coef std err t P>|t| [0.025 0.975]
------------------------------------------------------------------------------
Intercept 3331.8940 22.218 149.961 0.000 3288.294 3375.494
fitted -53.2676 2.034 -26.192 0.000 -57.258 -49.277
==============================================================================print(first_stage.summary().tables[1])==============================================================================
coef std err t P>|t| [0.025 0.975]
------------------------------------------------------------------------------
Intercept 42.0637 0.340 123.751 0.000 41.397 42.731
price -5.0058 0.053 -95.219 0.000 -5.109 -4.903
==============================================================================Rによるデータ演習
proxdata = pd.read_csv('proximity.csv')
proxdata.head()| lwage | educ | exper | black | south | smsa | nearc4 | |
|---|---|---|---|---|---|---|---|
| 0 | 6.306275 | 7 | 16 | 1 | 0 | 1 | 0 |
| 1 | 6.175867 | 12 | 9 | 0 | 0 | 1 | 0 |
| 2 | 6.580639 | 12 | 16 | 0 | 0 | 1 | 0 |
| 3 | 5.521461 | 11 | 10 | 0 | 0 | 1 | 1 |
| 4 | 6.591674 | 12 | 16 | 0 | 0 | 1 | 1 |
formula = 'lwage ~ 1 + educ + exper + black + south + smsa'
OLS = IV2SLS.from_formula(formula, data = proxdata).fit(cov_type='unadjusted')
print(OLS.summary.tables[1]) Parameter Estimates
==============================================================================
Parameter Std. Err. T-stat P-value Lower CI Upper CI
------------------------------------------------------------------------------
Intercept 4.9133 0.0631 77.917 0.0000 4.7897 5.0369
black -0.1882 0.0178 -10.604 0.0000 -0.2230 -0.1534
educ 0.0738 0.0035 20.908 0.0000 0.0669 0.0807
exper 0.0393 0.0022 17.924 0.0000 0.0350 0.0436
smsa 0.1647 0.0157 10.509 0.0000 0.1340 0.1955
south -0.1291 0.0152 -8.4829 0.0000 -0.1589 -0.0992
==============================================================================formula = 'lwage ~ 1 + exper + black + south + smsa + [educ ~ nearc4]'
TSLS = IV2SLS.from_formula(formula, data = proxdata).fit(cov_type='unadjusted')
print(TSLS.summary.tables[1]) Parameter Estimates
==============================================================================
Parameter Std. Err. T-stat P-value Lower CI Upper CI
------------------------------------------------------------------------------
Intercept 3.9398 0.8309 4.7414 0.0000 2.3112 5.5684
black -0.1296 0.0532 -2.4357 0.0149 -0.2339 -0.0253
exper 0.0623 0.0197 3.1663 0.0015 0.0237 0.1008
smsa 0.1348 0.0303 4.4555 0.0000 0.0755 0.1941
south -0.1093 0.0232 -4.7186 0.0000 -0.1546 -0.0639
educ 0.1318 0.0495 2.6624 0.0078 0.0348 0.2289
==============================================================================print(compare({'OLS':OLS, '2SLS':TSLS}, precision = 'std-errors', stars = True)) Model Comparison
==================================================
OLS 2SLS
--------------------------------------------------
Dep. Variable lwage lwage
Estimator OLS IV-2SLS
No. Observations 3010 3010
Cov. Est. unadjusted unadjusted
R-squared 0.2788 0.2140
Adj. R-squared 0.2776 0.2127
F-statistic 1163.4 673.47
P-value (F-stat) 0.0000 0.0000
================== ============ ============
Intercept 4.9133*** 3.9398***
(0.0631) (0.8309)
black -0.1882*** -0.1296**
(0.0178) (0.0532)
educ 0.0738*** 0.1318***
(0.0035) (0.0495)
exper 0.0393*** 0.0623***
(0.0022) (0.0197)
smsa 0.1647*** 0.1348***
(0.0157) (0.0303)
south -0.1291*** -0.1093***
(0.0152) (0.0232)
==================== ============== ==============
Instruments nearc4
--------------------------------------------------
Std. Errors reported in parentheses