Replication for “Training Microentrepreneurs over Zoom: Experimental evidence from Mexico” by Elwyn Davies, Peter Deffebach, Leonardo Iacovone, and David McKenzie

Overview

The code in this folder generates all tables and figures in the paper Training Microentrepreneurs over Zoom: Experimental evidence from Mexico” by Elwyn Davies, Peter Deffebach, Leonardo Iacovone, and David McKenzie. The replicator should expect the code to run for 20 minutes. Downloading the necessary R packages may take additional time.

Data Accessibility Statement

Data sources

There are three data sources used in this paper

1. Original newly collected data from our experiment. Data file: data/merged_all_surveys.dta

2. Data from the baseline survey of a 2018 experiment with CREA. See here for a link to the published article an the AER P&P. Replication data available here. We certify that we have permission to distribute the 2018 CREA data. However for simplicity we only provide variables used in the analysis, Table 1 of the main paper, which shows summary statistics of previous CREA trainees to better compare to our sample. The .do files which clean the raw data are provided, but cannot be run. Data file: data/crea_2018_cleaned.dta

3. Data from the 1st quarter 2023 National Survey of Occupation and Employment in Mexico (ENOE). Download the data here. We do not have express permission to distribute the data ourselves. For simplicity we only provide variables used in the analysis, Table 1 of the main paper, which shows summary statistics of women entrepreneurs in Mexico to better compare to our sample. The .do files which clean the raw data are provided, but cannot be run. Data file: data/enoe_cleaned.dta

License for data

The code is licensed under a Creative Commons/CC-BY-NC/CC0 license. See license.txt for details. The original survey data (data/merged_all_surveys.dta) is also licensed uncer a a Creative Commons/CC-BY-NC/CC0 license. For other data used in this project, consult the original sources.

Summary of data

Description of folder contents

This folder structure of this document is as follows:

• ado: Holds the Stata files

• data: Holds the cleaned dataset on women entrepreneurs in Mexico and Guatemala. Also holds the 2018 CREA data and the ENOE data.

• out: Will store all the tables and figures produced in the analysis.

• out_example: Stores the tables and figures produced by the code when I ran it. These match the tables and figures in the paper. It will be useful for checking results match.

• src: Holds all code used in the production of the tables and figures. See instructions below for more details.

• tex: Contains the LaTeX documents for compiling the results in the main paper and the pre-analysis plan.

• training_session_data: Contains additional data on the content of the training sessions.

Computational requirements

Software requirements

Stata and R are used in the analysis of this project. This project was written using Stata 18 and has not been tested on older versions. I recommend using the latest version of R, 4.3.1, to run the R code. As described below, we use renv to keep track of dependencies, and only the exact versions of dependencies are listed. As a consequence, if a package listed does not support older versions of R, there may be issues instantiating the environment.

Package dependencies

The ado folder lists all dependencies for the Stata portion of the code. The renv.lock file lists all dependencies for the R portion of the code.

Deterministic randomness

Seeds are set at

• Line 9 of src/heterogeneous_effects/run_generic_ml.R
• Line 105 of src/heterogeneous_effects/define_get_churn.R
• Line 9 of src/heterogeneous_effects/zaminfluence_mexico.R

Instructions for replication

Note: I do some post-processing in Julia to clean up significant figures for the tables. For example, I round sales outcomes, which are in the thousands, to the nearest integer. Any differences between the tables produced in this folder and what appears in the working paper should be due to this rounding.

Instructions for replicating tables and figures made in Stata:

1. Open Stata.

2. Set the working directory to the one this file lives in. There is no other path to set. We will never change directories during this code.

3. We record all dependencies in Stata using the ado folder and by adding this folder to the front of the list of ado paths in Stata. This means you should not need to use ssc at all. If something pops up that you need to download, please download it. Note: you should not need to download anything to use the LaTeX table maker frmttable. But if you encounter an error, do not download frmttable via ssc. The frmttable command is exported by outreg, so you should download outreg

4. Run do main.do and all code that uses Stata (to create the vast majority of tables and figures) will run. This is the only do-file you should need to run.

Instructions for replicating tables and figures made in R

Replicating all tables and figures except for Table B.5

The following are instructions to make a tables and figures produced in the R scripts except the ZAM Influence table (Table B.5). As discussed below, we use renv to keep track of dependencies. But since the zaminfluence package is not registered on CRAN, we couldn’t automate the downloading process.

1. Open R.

2. Again, set the working directory to the one this file lives in. If you open R in this directly, renv might bootstrap itself and install and activate.

3. We record all the dependencies in R using the R package renv. Please download renv using install.packages("renv") and then run renv::activate() and renv::restore(). R should now download all the necessary dependencies.

4. Run source("main.R"). This will take a long time to run, as the GenericML machine learning process is computationally heavy.

Replicating Table B.5

To replicate Table B.5, we are going to install the zaminfluence package without using renv. There is a separate step here because it very hard to install this package.

To install the the zaminfluence package, run

renv::deactivate()

install.packages("devtools") # Skip if already installed
library(devtools)
devtools::install_github("https://github.com/rgiordan/zaminfluence/",
                           ref="master",
                           subdir="zaminfluence",
                           force=TRUE)

install.packages("haven") # Skip if already installed
install.packages("tidyverse") # Skip if already installed
install.packages("fixest") # Skip if already installed
install.packages("gridExtra") # Skip if already installed

To create Table B.5, run

source("src/heterogeneous_effects/zaminfluence_mexico.R")

Remember to run renv::activate() after finishing to re-load the package environment.

List of Tables and Figures

Instructions for compiling the pdf

Using Latex, compile the documents paper.tex and pap.tex. I have included paper_example.tex and pap_example.tex which look in the folder out_example for tables and figures. Use these to debug any LaTeX compilation errors.