OLS

import matplotlib
if not hasattr(matplotlib.RcParams, "_get"):
    matplotlib.RcParams._get = dict.get

OLS#

from linearmodels.datasets import wage_panel

data = wage_panel.load()
data.head(10)

# nr: individual id
# lwage: log of wage -> y
# married -> T
nr year black exper hisp hours married educ union lwage expersq occupation
0 13 1980 0 1 0 2672 0 14 0 1.197540 1 9
1 13 1981 0 2 0 2320 0 14 1 1.853060 4 9
2 13 1982 0 3 0 2940 0 14 0 1.344462 9 9
3 13 1983 0 4 0 2960 0 14 0 1.433213 16 9
4 13 1984 0 5 0 3071 0 14 0 1.568125 25 5
5 13 1985 0 6 0 2864 0 14 0 1.699891 36 2
6 13 1986 0 7 0 2994 0 14 0 -0.720263 49 2
7 13 1987 0 8 0 2640 0 14 0 1.669188 64 2
8 17 1980 0 4 0 2484 0 13 0 1.675962 16 2
9 17 1981 0 5 0 2804 0 13 0 1.518398 25 2

1. OLS with Control: Ceteris Paribus – to control what we can see#

import statsmodels.formula.api as smf

ols_model = smf.ols('lwage ~ expersq + union + married + hours', data=data).fit()
ols_model.summary().tables[1]
coef std err t P>|t| [0.025 0.975]
Intercept 1.5327 0.032 47.842 0.000 1.470 1.595
expersq 0.0012 0.000 6.208 0.000 0.001 0.002
union 0.1679 0.018 9.239 0.000 0.132 0.204
married 0.1966 0.016 11.948 0.000 0.164 0.229
hours -3.33e-05 1.42e-05 -2.352 0.019 -6.11e-05 -5.54e-06

2. Using Fixed Effect – to control what we cannot see#

2-1. Entity Effects (time-fixed confounders)#

from linearmodels.panel import PanelOLS
mod = PanelOLS.from_formula("lwage ~ expersq + union + married + hours + EntityEffects",
                            data=data.set_index(["nr", "year"]))

result = mod.fit(cov_type='clustered', cluster_entity=True)
result.summary.tables[1]
Parameter Estimates
Parameter Std. Err. T-stat P-value Lower CI Upper CI
expersq 0.0040 0.0002 16.552 0.0000 0.0035 0.0044
union 0.0784 0.0236 3.3225 0.0009 0.0322 0.1247
married 0.1147 0.0220 5.2213 0.0000 0.0716 0.1577
hours -8.46e-05 2.22e-05 -3.8105 0.0001 -0.0001 -4.107e-05

2-2. Time Effects (entity-fixed confounders)#

from linearmodels.panel import PanelOLS

mod = PanelOLS.from_formula("lwage ~ expersq + union + married + hours + TimeEffects",
                            data=data.set_index(["nr", "year"]))

result = mod.fit(cov_type='clustered', cluster_time=True)
result.summary.tables[1]
Parameter Estimates
Parameter Std. Err. T-stat P-value Lower CI Upper CI
expersq -0.0021 0.0002 -10.845 0.0000 -0.0025 -0.0017
union 0.1721 0.0210 8.2153 0.0000 0.1311 0.2132
married 0.1634 0.0070 23.221 0.0000 0.1496 0.1772
hours -6.535e-05 3.629e-05 -1.8010 0.0718 -0.0001 5.789e-06

2-3. Use Both Entity and Time Effects#

from linearmodels.panel import PanelOLS

mod = PanelOLS.from_formula("lwage ~ expersq + union + married + hours + EntityEffects + TimeEffects",
                            data=data.set_index(["nr", "year"]))

result = mod.fit(cov_type='clustered', cluster_entity=True, cluster_time=True)
result.summary.tables[1]
Parameter Estimates
Parameter Std. Err. T-stat P-value Lower CI Upper CI
expersq -0.0062 0.0008 -8.1479 0.0000 -0.0077 -0.0047
union 0.0727 0.0228 3.1858 0.0015 0.0279 0.1174
married 0.0476 0.0177 2.6906 0.0072 0.0129 0.0823
hours -0.0001 3.546e-05 -3.8258 0.0001 -0.0002 -6.614e-05

질문이나 의견을 남겨주세요.#

Contents