JMSLab Template

A template for research projects developed by JMSLab.

Prerequisites

• We recommend installing Anaconda 3.8 or above to get Python.
  • We use Python to run SCons; custom SCons builders are also written in Python.
  • You will need to install the dependencies in source/lib/requirements.txt (including SCons). See quick start below.
• git for version control.
  • And git-lfs for versioning large files.
• LyX for writing documents.

In addition, each project may use other specialized tools. For the working example in this template, install:

• R
• Stata
• Matlab

Repository structure

• source/ contains source scripts and (small) raw data.

• output/ and datastore/ should mimic the folder structure in source/.

  • For instance, the code in source/analysis/plots/ saves output to output/analysis/plots/.

  • datastore/ is not under version control; a project's large files are stored here. See here for an example designed to work with this template.

  • With the exception of large raw files, every file in these folders is produced by source/.

  • Files tracked by git-lfs are ignored in output/ because git-lfs files are not permitted in a template repository.

• temp/ is not under version control; create the folder after cloning the repository. temp/ may used by scripts to store temporary or intermediate files (though some projects may not need it).

• Issue folders: When working on issue branches, you may create an issue folder at the top of the directory under version control (e.g. ./issue1_short_task_name).

  • Code and (small) deliverables related to the issue are organized inside the issue folder. See this example.

  • The issue folder should be deleted after the issue is resolved and is not merged into the main branch.

• /.github defines workflows for github actions. We recommend disabling these by default when creating a private repository.

Quick start

1. Open the command-line and clone the repository. For example,

   git clone https://github.com/JMSLab/Template ProjectName
   cd ProjectName

2. Create a symbolic link called datastore to a local copy the project's datastore, if one exists for the project (e.g. a Dropbox or Google Drive folder).

   • Do not link a "live" copy of the datastore (i.e. one that is synchronized to the internet). Work with a local, off-line copy before modifying the live copy; otherwise the data may get unintentionally overwritten for everyone using the datastore.

3. Install dependencies:

   pip install -r source/lib/requirements.txt

   (If using conda, run conda install --file source/lib/requirements.txt.) Requirements for other languages, should there be any, will be found in source/lib/requirements.{ext} with {ext} equal to do (Stata), r (R), m (Matlab), and so on.

4. Make sure that all the required program executables are in your system's path.

   • The default names that SCons assumes for the programs are in source/lib/JMSLab/builders/executables.yml.

   • To have SCons use a custom executable name or path, define a command-line (environment) variable named JMSLAB_EXE_PROGRAM. e.g. On Windows, SET JMSLAB_EXE_STATA=StataSE.exe or SET JMSLAB_EXE_STATA=C:\Program Files (x86)\Stata16\StataSE.exe.

5. To compile the project, open the command-line and run scons from the project's root folder.

   • To run a given script or create a given file, run scons path/to/script_or_file; this will recursively run all the dependencies required to run the script or create the file. e.g. scons output/derived/wb_clean/gdp_education.csv.

SConscript files

In order to integrate a new script into the SCons build, you need to modify the SConscript file in the corresponding source/ sub-folder. For example, to add source/derived/wb_clean/takelogs.do to the SCons build, add an entry to source/derived/SConscript. In this case:

target = ['#output/derived/wb_clean/gdp_education_logs.csv']
source = ['#source/derived/wb_clean/takelogs.do',
          '#output/derived/wb_clean/gdp_education.csv']
env.Stata(target, source)

• target is a list with all of the files produced by the script.

• source is a list with the script's name and all of the files used as input; the script must be the first element of the list.

• env.Stata is the Stata builder provided in source/lib/JMSLab/builders; this is imported and saved as part of the env object in the SConstruct file at the root of the project.

• For tips on batch-specifying targets and sources, see ./docs/batch_specifying.md.

Citations and expectations for usage

This template is based on gslab-econ/Template/v4.1.3 and gslab-python/v4.1.4.

It was designed for use in JMSLab. You are welcome to use it in your own research, as long as you (i) let us know that you plan to do this and (ii) agree to help maintain the template in the future.

Use in reproducible research

A research repository based on this template was used successfully in a first-try replication by a third party contracted to the American Economic Association in June 2022.