Mass Media Field Experiment Archive for Replication

This project is to provide dataset and analysis files for A Mass Media Experiment in Rural Uganda to violence against women by encouraging disclosure. The published paper is here. This multi-wave Randomized Controlled Trials (RCTs) have been conducted by Donald Green, Anna Wilke, and Jasper Cooper. I was responsible for building this archive repository to make this large-scale experiment dataset more accessible to academic commons and the general public.

Affiliation: Institute for Social and Economic Research and Policy, Columbia University

Keywords: Randomized Controlled Trials, Replication Archive

Software: R, LaTex, bash

This repository is also published on Harvard Dataverse

How to run the code

0. Install the required packages

   #!/usr/bin/bash
   while IFS=" " read -r package version; 
   do 
     Rscript -e "devtools::install_version('"$package"', version='"$version"')"; 
   done < "requirements.txt"

1. Open PV_replication.Rproj to ensure that all file paths are set relative to the replication archive

2. Open __main_script.R and run all scripts from here

3. True / false logics switch on and off scripts that take a long time to run

Content Menu

• 01_data/ = Contains raw data
  • boostrap_extrapolation.Rdata = Stored output extrapolation
  • cluster_level_data/ = Village-level data
    • film_festival.csv = Data on intervention
    • location_data.csv = Data that link cluster IDs to districts
    • sampling_radius.csv = Radii for sampling of Rs
    • treatment_assignment.csv = Random assignment
  • codebooks/ = ODK tablet coding
    • endline_choices.csv = Endline survey answer choices
    • endline_Qs.csv = Endline survey questions
    • midline_choices.csv = Midline survey answer choices
    • midline_Qs.csv = Midline survey questions
    • vht_endline_choices.csv = VHT endline survey answer choices
    • vht_endline_Qs.csv = Endline VHT survey questions
    • vht_midline_choices.csv = VHT midline survey answer choices
    • vht_midline_Qs.csv = Midline VHT survey questions
  • household/ = Household level data
    • distance_data.csv = Distance of Rs to video hall
    • endline_1.csv = Endline HH survey
    • endline_2.csv = Endline HH survey
    • endline_3.csv = Endline HH survey
    • endline_4.csv = Endline HH survey
    • midline_1.csv = Midline HH survey
    • midline_2.csv = Midline HH survey
    • midline_3_1.csv = Midline HH survey
    • midline_3_2.csv = Midline HH survey
    • midline_3.csv = Midline HH survey
    • midline_4.csv = Midline HH survey
    • midline_5.csv = Midline HH survey
    • random_sampling_1.csv = Random sampling of Rs in midline
    • random_sampling_2.csv = Random sampling of Rs in midline
  • lasso_selected_covariates.Rdata = Covariates selected through lasso
  • UG_VAW_data.Rdata = Stored full data
  • vht/ = VHT datasets
    • vht_el_1.csv = VHT endline survey
    • vht_ml_1.csv = VHT midline survey
    • vht_ml_2.csv = VHT midline survey
    • vht_ml_3.csv = VHT midline survey
    • vht_ml_4.csv = VHT midline survey
• 02_code/ = Code scripts
  • __main_script.R = Main script that runs others
  • 00_useful_functions/ = Functions used throughout
  • 01_codebook/ = Scripts that build codebooks
  • 02_load_and_clean_data/ = Loading and cleaning datasets
  • 03_variable_coding/ = Coding outcomes and covariates
  • 04_merging/ = Merging datasets
  • 05_covariate_selection/ = Lasso covariate selection
  • 06_analyses_paper/ = Analysis scripts
  • 07_analyses_appendix/ = Robustness check scripts
• 03_tables/ = Tables are output here
• 04_figures/ = Figures are output here
• IPV_replication.Rproj = Run everything from here

Notes

• Most analyses are based on a panel of "compliers" interviewed in both the midline and endline surveys. This subset can be identified using the subset respondent_category == "Complier".
• We did not re-ask all questions of those in the panel, so their endline responses are merged in from midline.
• The multiple versions of the raw data correspond to the different datasets output by ODK / CSO when a change is made to the survey. Each change requires a new survey version, thus producing a new dataset.
• The only change made to the raw data files is the removal of PII and variables that are not used in the analysis.
• All other modifications made to data (changing of values, etc.) in cleaning scripts were implemented by field manager over course of field work.