Reproducibility package for CLARE: A Causal Machine Learning Approach To Resilience Estimation

To reproduce the findings in this paper, new users should follow the steps below:

1. Descriptive Statistics (Stata): Open STEP 0 - Descriptive_Statistics.do, update the file path in line 5, and run the code.
2. Estimation and Forecasting (R)
   • Open the .Rproj file included in the package.
   • Restore the environment using renv::restore() (recommended), or manually install the required packages.
   • Run the script STEP 1 - Estimation_Forecasting.R.
   • Then run STEP 1 - Estimation_Main_Country.R, which executes all estimation steps for Malawi, Nigeria, Tanzania, and Uganda.
3. Aggregation and Evaluation – Forecasting (Stata)
   Open STEP 2 - Aggregation and Evaluation_Forecasting.do, update the file path in the indicated line, and run the code.
4. Aggregation and Evaluation – Out-of-Country (Stata): Open STEP 2 - Aggregation and Evaluation_Out-Of-Country.do, update the file path in the indicated line, and run the code.
5. Figures (Stata): Open STEP 3 - Figures do file, update the file path in the indicated line, and run the code.

Note: The full workflow of the package can be executed and reproduces all results reported in the paper. However, the starting point of the reproducibility package is intermediate data, as the raw LSMS-ISA household microdata were merged at an early stage of the analysis with confidential household geographic coordinates, which are restricted and cannot be shared.

For transparency, the complete set of scripts used by the authors to process the raw data and construct the intermediate datasets is included in the folder Codes/Do files - Variable Construction. These scripts were not executed by the replicators, as access to the raw confidential data is restricted. Instead, the replicators virtually verified the data construction process by reviewing the code and confirming that the outputs generated by the authors correspond exactly to the intermediate datasets provided as the starting point of this package. The included intermediate datasets are sufficient to fully reproduce all analyses, tables, and figures presented in the paper.