*******************************
BLIND ANALYSIS REPLICATION CODE
*******************************

Third-Party Performance Pay to Improve Local Government Accountability: A Field Experiment in Burkina Faso

Authors: 
Malte Lierl (German Institute for Global and Area Studies), malte.lierl@giga-hamburg.de (*)
Marcus Holmlund (World Bank), mholmlund@worldbank.org
(*) corresponding author and PI

Code contributors: Kamil Kouhen, Malte Lierl

Date: 2022-11-07

**Purpose**

This package contains the code for the results-blind analysis of the study "Third-Party Performance Pay to Improve Local Government Accountability: A Field Experiment in Burkina Faso". To obtain the unblinded results, same code should be re-run on the unblinded data, after registration of the blind analysis report and unblinding of the study. This blind analysis has been conducted prior to the authors' access to the actual results of the experiment. 

This package contains the blind analysis of outcome data. It does not contain the analysis of covariate balance at baseline, treatment compliance, attrition and nonresponse, which are also included in the blind analysis report. The latter analyses use different datasets which have been separated from the blinded outcome data. 


*********
CONTENTS
*********
 
The root folder contains a script to choose between blinded and unblinded analysis modes ('00A - CHOOSE BLINDED or UNBLINDED'), which is to be run prior to running the master script ('00B - MASTER'). 

**Code**
The sub-folder 'Code' contains all other code for user-written functions, data cleaning and preparation, generation of mock treatment identifiers (blind analysis only), and analysis. All scripts in this folder are sourced by the master script. 

**Data**
The sub-folder 'Data' contains the datasets loaded by the master script:  
- A sub-folder 'Blinded' contains the data for the blind analysis. 
- To run the unblinded analysis, a sub-folder 'Unblinded' should be added that contains the unblinded data, but otherwise mirrors the structure of the 'Blinded' data folder. 

Each data sub-folder ('Blinded'/'Unblinded') is structured into three folders: 'Source', 'Intermediate', 'Final'. The datasets in 'Intermediate' and 'Final' are generated as part of the data cleaning and preparation process. All source data are in the 'Source' folder. After data  cleaning and construction of outcome variables, the datasets in the 'Final' folder are used for analysis. 

**Supporting Documents**
This folder contains original questionnaires and documentation regarding the endline data collection process (1_ field procedure manual, 2_ surveyor manuals, 3_ in-person questionnaires, 4_ phone-based questionnaires, 5_ implementation reports, 6_ codebooks). 

**Output**
This folder contains the output of the blind analysis. All outputs used in the blind analysis report are written into the sub-folder 'For Publication'. The sub-folder 'For Cleaning' contains now-obsolete intermediate outputs that were generated as part of the data cleaning process. The sub-folder 'Documents' contains a list that details some of the judgment calls that were made during the blind data cleaning and analysis process. 

*********
IMPORTANT
*********

The source data is not included with the blind analysis code package uploaded to the AEA and/or OSF registries. If data access is needed for the purpose of replicating this blind analysis prior to the publication of the study, please kindly direct a request to the authors. Due to the ongoing conflict in Burkina Faso, appropriate safeguards will have to be taken for the protection of human subjects and of sensitive information. 



