Introduction to Effectiveness Verification-Causal Reasoning for Correct Comparison / Basics of Econometrics Reproduce the source code in Python To do.
I already have a Great ancestor implementation example, but I will leave it as a memo for my study.
This article describes Chapters 4 and 5. The code is also posted on github. In addition, variable names and processing contents are basically implemented in the book.
CausalImpact
Here, it is implemented using pycausal impact.
The result does not match the R version, but this seems to be because the method handled is different as described in the README.
It seems that you can specify the state space model with the argument model, but I gave up because it seems to be difficult to match the implementation with the R version.
CausalImpact
from causalimpact import CausalImpact
impact = CausalImpact(CI_data, pre_period, post_period, prior_level_sd=None)
impact.plot()
print(impact.summary())
RDD Non-linear regression analysis can be performed by learning the following model and referring to the effect size of the intervention variable.
RDD
from sklearn.preprocessing import PolynomialFeatures
degree = 4
X = nonlinear_rdd_data[['history_log']]
X = X - cutpoint
X_poly = PolynomialFeatures(degree=degree, include_bias=False).fit_transform(X)
X_poly = pd.DataFrame(X_poly, columns=[f'X{i+1}' for i in range(X_poly.shape[1])])
nonlinear_rdd_data = pd.concat([nonlinear_rdd_data, X_poly], axis=1)
nonlinear_rdd_ord4 = ols('visit ~ treatment + X1 + X2 + X3 + X4 + treatment:X1 + treatment:X2 + treatment:X3 + treatment:X4', data=nonlinear_rdd_data).fit()