A Reproducible Workflow for Economics Research Using Snakemake and R

Introduction

We provide a template for a reproducible research project using Snakemake and the R programming language.
We use Snakemake to construct a set of rules that form a DAG that implements the entire research pipeline, starting with some data cleaning, running some regressions, constructung figures and tables, and then finishing with compiling a pdf article and slides.
We believe this mimics an approximate workflow of most empirical research in economics.
The R language is used to perform all steps of our analysis and to compile our pdf documents (the latter via the knitr, bookdown and rticles packages).
Snakemake allows us to construct a set of interconnect rules to build our workflow from start to finish - and to partially update the build where required when we update our scripts, data and parameters.

Motivating Example

Our example project involves replicating the main tables and figures of Mankiw, Romer and Weil’s classic 1992 QJE article “A Contribution to the Empirics of Economic Growth.”
We hope by using an example that is simple in its methods readers focus on how we have chosen to assemble both pure R codes and the Snakemake rules that build our project, rather than getting lost on econometric methodologies.

Installation Instructions

Follow these Steps to install the necessary software on your system

You need to have the following software and packages installed:

1. Python 3 (Python 3.6 or higher)
2. Snakemake (we’ll install the correct version in a couple of lines time!)
3. R (version 4.0.x)
   • For R version 3.6.x see an archived version here.

Installing Python

Either:

1. Install Anaconda Python:
   • We provide instructions on how to install anaconda python here
2. Install Python using the deadsnakes ppa:
   • Here’s how to add the deadsnakes ppa and install Python 3.8

   $ sudo apt-get install software-properties-common
   $ sudo add-apt-repository ppa:deadsnakes/ppa
   $ sudo apt-get update
   $ sudo apt-get install python3.8
   

Installing Snakemake

We have included a requirements.txt file that we can use to install a specific version of snakemake.
This makes sure that our example runs on your machine (or at least won’t break because you use a different version of snakemake than we do)

pip3 install -r requirements.txt

you may need to replace pip3 with pip

Note: In recent versions of Snakemake you must state how many cores you want to use when executing rules.
In this example, we encourage the use of a single core, so when we use snakemake we will always write it as snakemake --cores 1 <something>.
More cores speeds things up, but our example is quite light so it seems unnecessary.

Installing R

We provide instructions on how to install R here

Installing pandoc and LaTeX

Because we want to generate pdf outputs we need two additional bits of software to make that happen:

• If you do not have RStudio installed, you will have to install Pandoc (http://pandoc.org)
• If you do not have LaTeX installed, we recommend that you install TinyTeX (https://yihui.name/tinytex/)
  • TinyTeX is a lightweight, portable, cross-platform, and easy-to-maintain LaTeX distribution.
  • From inside R:

    install.packages('tinytex')
    tinytex::install_tinytex()  # install TinyTeX
    

Install the Required R libraries

We utilize many additional R packages inside the scripts that build our project.
To ensure that our project runs on every machine without issues relating to R packages not being installed we utilize renv to control the list of packages needed to run this example, and to monitor the version of the package we use.

Once you have completed the installation instructions above, we have provided a simple command to install renv.
Open a terminal and navigate to this directory.
Then in the terminal enter the following command to install renv:

snakemake --cores 1 renv_install

Then you will need to provide consent for renv to be able to write files to your system:

snakemake --cores 1 renv_consent

Once this is complete you can use renv to create a separate R environment that contains the packages we use in our example by entering the following command into the terminal:

snakemake --cores 1 renv_init

The above command will initialize a separate R environment for this project.

Now we will install the necessary packages (and their precise versions) which are stored in the renv.lock file:

snakemake --cores 1 renv_restore

This will install all the packages we need. It may take a while.

Running the Example

Once the installation instructions are complete, we can run the project.
The result will be 2 pdfs: (1) mrw_replication.pdf - a pdf with some figures and tables of results included; and
(2) mrw_replication slides - a beamer slide deck that also contains some figures and pdfs that were produced by the workflow.

To run the project, enter the following into the terminal:

snakemake --cores 1 all

This will run through all the R scripts in order to complete the build of the project.

Visualization of the Workflow

Snakemake workflows are a directed acyclic graph (DAG).
We can visualize the relationship between the rules (a simplified view of the DAG) in our workflow:

Check out the rules in ./rules/dag.smk for various visualizations of the workflow.
You will need to install graphviz to run these rules - we have included a rule inside dag.smk to install this for you.

Archiving R packages in a new project

• If you are starting a new project, which is likely if you are using
  this template, you need to initialize a new renv instance to
  track your R packages and store them.
  Enter the following command into the same terminal as above
  and press Return.

  snakemake --cores 1 renv_init
  

• If we add new R packages that we want to include in a project, we take a snapshot of the packages utilized in the project with:

snakemake --cores 1 renv_snap

Updates to Workflow Example

We plan to periodically update the workflow as we find better/simpler ways to do things and as our opinions on best practice evolve.
Major changes are tracked in the NEWS file with brief descriptions of the changes implemented.

Comments / Suggestions / Issues

We’d love to hear your comments, suggestions or installation issues encountered when running the example.
Post an issue on Github.

Suggested Citation

Deer, Lachlan and Julian Langer and Ulrich Bergmann. 2021. A Reproducible Workflow for Economics Research Using Snakemake and R.