理解双重差分

[chengjun]

双重差分本质就是treated和time两个dummy变量之间的调节作用

obs: 820 Dataset from Card&Krueger (1994)

• id int Store ID
• t byte Feb. 1992 = 0; Nov. 1992 = 1
• treated long treated New Jersey = 1; Pennsylvania = 0
• fte float Output: Full Time Employment
• bk byte Burger King == 1
• kfc byte Kentuky Fried Chiken == 1
• roys byte Roy Rogers == 1
• wendys byte Wendy's == 1

import pandas as pd
import statsmodels.formula.api as sm
import statsmodels.stats.sandwich_covariance as sw
import numpy as np
import statsmodels as statsmodels

df = pd.read_stata('data/cardkrueger1994.dta', 
                    convert_categoricals=False, convert_missing=False)

model = sm.ols('fte~t + treated + t*treated + bk + kfc+ roys + wendys', data = df).fit() # (cov_type='HC1',)
model.summary()

Index(['id', 't', 'treated', 'fte', 'bk', 'kfc', 'roys', 'wendys'], dtype='object')

import matplotlib.pyplot as plt
from statsmodels.graphics.factorplots import interaction_plot

plt.style.use('ggplot')
fig = interaction_plot(
   x        = df['t'],
   trace    = df['treated'],
   response = model.fittedvalues,
   colors   = ['red','blue'],
   markers  = ['D','^'])
plt.xlabel('Time')
plt.ylabel('FTE')
plt.legend().set_title('Treated')
plt.show()

# check by calculating the mean for each group directly
contro_dif = df[(df.treated == 0)&(df.t==1)]['fte'].mean() - df[(df.treated == 0)&(df.t==0)]['fte'].mean()
treated_dif = df[(df.treated == 1)&(df.t==1)]['fte'].mean() - df[(df.treated == 1)&(df.t==0)]['fte'].mean()

treated_dif - contro_dif

2.913982391357422

[chengjun]

https://github.com/SocratesAcademy/css/blob/master/notebook/more/03-06-difference-in-difference.ipynb stata version

[chengjun]

The Missing Women: Another Case of DID #36