Reproducing Results from A Hybrid Approach to Targeting Social Assistance

Introduction

This repository contains the code and data required to reproduce the results found in "A Hybrid Approach to Targeting Social Assistance". Specifically, to run simulation studies that estimate out of sample error rates using the Hybrid, Hybrid-AI, Hybrid-EC, and Hybrid-DU models on data from Indonesia (Alatas et al. (2012)) and Burkina Faso (Hillebrecht et al. (2020)).

Requirements

To install the required R packages, run the following code in R:

install.packages(c("truncnorm", "mvtnorm", "LaplacesDemon", "MASS", "dplyr",
                   "ggplot2", "Rcpp", "reshape2", "caret", "parallel"))

Data

We use two sources of data containing community based rankings, survey information, and consumption/expenditure data. This data can be found in the following sub-directories:

list.files("Data/Burkina Faso/Cleaning/")

## [1] "cleaning.do"              "hillebrecht.csv"          "hillebrecht.dta"         
## [4] "hillebrecht(missing).csv" "hillebrecht(missing).dta" "variables.csv"

list.files("Data/Indonesia/Cleaning/")

##  [1] "alatas.csv"                               
##  [2] "alatas.dta"                               
##  [3] "alatas(missing).csv"                      
##  [4] "alatas(missing).dta"                      
##  [5] "cleaning.do"                              
##  [6] "FAO Dietary Diversity Guidelines 2011.pdf"
##  [7] "food.dta"                                 
##  [8] "notes.docx"                               
##  [9] "ranks.dta"                                
## [10] "variables.csv"                            
## [11] "xvars.dta"

The data files that will be called are "hillebrecht.csv" and "alatas.csv".

Reproduce

1. Run run_simulations.R to reproduce error rate results and coefficient estimate results.

• Indonesia Analysis/all_results.csv
• Indonesia Analysis/all_coef.csv
• Indonesia Analysis/coef_total_sample.csv
• Indonesia Analysis/CB_beta_rank_CI_noelite.csv
• Indonesia Analysis/CB_beta_rank_CI.csv
• Burkina Faso Analysis/all_results.csv
• Burkina Faso Analysis/all_coef.csv
• Burkina Faso Analysis/coef_total_sample.csv
• Burkina Faso Analysis/CB_beta_rank_CI_noelite.csv
• Burkina Faso Analysis/CB_beta_rank_CI.csv

The above files can be used to generate plots found in the manuscript:

2. Run Burkina Faso Analysis/make_plots.R to reproduce error rate plots and coefficient plots for the Burkina Faso data.

• Burkina Faso Analysis/coef_score_EC_hillebrecht.pdf
• Burkina Faso Analysis/coef_score_hillebrecht.pdf (Figure 1)
• Burkina Faso Analysis/ER_hybrid_AI.pdf (Figure 7 a)
• Burkina Faso Analysis/ER_hybrid_DU.pdf (Figure 8)
• Burkina Faso Analysis/ER_hybrid.pdf (Figure 3 a)

3. Run Indonesia Analysis/make_plots.R to reproduce error rate plots and coefficient plots for the Indonesia data.

• Indonesia Analysis/coef_score_EC_hillebrecht.pdf (Figure 5)
• Indonesia Analysis/coef_score_hillebrecht.pdf (Figure 2)
• Indonesia Analysis/ER_hybrid_AI.pdf (Figure 7 b)
• Indonesia Analysis/ER_hybrid_EC.pdf (Figure 6)
• Indonesia Analysis/ER_hybrid.pdf (Figure 3 b)

4. Run Burkina Faso Analysis/run_mcmc_weights.R to reproduce heterogeneous ranker results.

• Burkina Faso Analysis/heter_weights_omega.pdf (Figure 4 a)
• Burkina Faso Analysis/heter_weights_corr.pdf (Figure 4 b)

References

Alatas, V., Banerjee, A., Hanna, R., Olken, B., and Tobias, J. (2013).Targeting the poor: Evidence from a field experiment in Indonesia.Harvard Dataverse,https://doi.org/10.7910/DVN/M7SKQZ, V5.

Hillebrecht, M., Klonner, S., Pacere, N. A., and Souares, A. (2020b). Community-basedversus statistical targeting of anti-poverty programs: Evidence from Burkina Faso.Journalof African Economies, 29(3):271–305