scunning1975 · tomcaputo · Mar 7, 2021 · Mar 7, 2021 · Mar 7, 2021 · Mar 7, 2021
diff --git a/Python/Differences_in_Differences.ipynb b/Python/Differences_in_Differences.ipynb
diff --git a/Python/Directed_Acyclical_Graphs.ipynb b/Python/Directed_Acyclical_Graphs.ipynb
diff --git a/Python/Instrumental_Variables.ipynb b/Python/Instrumental_Variables.ipynb
diff --git a/Python/Matching_and_Subclassification.ipynb b/Python/Matching_and_Subclassification.ipynb
diff --git a/Python/Panel_Data.ipynb b/Python/Panel_Data.ipynb
diff --git a/Python/Potential_Outcomes.ipynb b/Python/Potential_Outcomes.ipynb
diff --git a/Python/Probability_and_Regression.ipynb b/Python/Probability_and_Regression.ipynb
diff --git a/Python/README.md b/Python/README.md
@@ -0,0 +1,9 @@
+# Notes on Python Implementation
+
+Code outputs have been verified and match R output except in the following cases:
+
+- **Differences_in_Differences.ipynb:** The design matrix in the Cunningham and Cornwell (2013) example is rank deficient. lm and lm_robust have a convergence issues. The problem seems to be with the fip variable. The statsmodels algorithm is more robust to rank deficiency resulting is different results
+- **Matching_and_Subclassification.ipynb:** Python does not have an implementation of nearest neighbor matching in python. I may be possible to recreate one use standard KNN tools
+- **Regression_Discontinuity.ipynb:** The smoothing and density section is missing
+
+The majority of the models use the statmodels package. While I tried to limit my usage of R code through rpy2 I use it once for synthetic control matching.
diff --git a/Python/Regression_Discontinuity.ipynb b/Python/Regression_Discontinuity.ipynb
diff --git a/Python/Synthetic_Control.ipynb b/Python/Synthetic_Control.ipynb