Skip to content

Latest commit

 

History

History
161 lines (108 loc) · 8.75 KB

README.md

File metadata and controls

161 lines (108 loc) · 8.75 KB
Raw

Citations

Testing for Causal Evidence on Data Sharing and Citations

See https://osf.io/cdt8y/ for more info.

To update the research log, click here.

Draft of the entire data collection process is here.

Folder Structure

The folder structure of this repository grew organically over the development of the project. There is also a Box folder URAPshared that contains several files not contained in this repository. The purpose of the Box folder is to manage the raw data from scraping and data entry. In the following, if a folder x is contained in the Box folder, then it will be denoted as URAPshared/x.

Octoparse

Data collection started with the journals AJPS and APSR and with the use of Octoparse. The folder octoparse contains the otd files to run this data collection. Octoparse uses otd files as instructions on how to collect data.

The output from the AJPS and APSR data collection is stored in URAPshared/octoparse in the following files

  • apsr_article_content_2006_2014.csv
  • apsr_issue_toc.csv
  • ajps_article_content_2003_2007.csv
  • ajps_article_content_2008_2012.csv
  • ajps_article_content_2013_2016.csv
  • ajps_issue_toc.csv

We collected article information from the page of each journal issue's table, and then separately collected article content from the individual article pages. This is what article_content and issue_toc refers to in the file names.

The collection with Octoparse has a few issues.

For AJPS, there are two issues. First, memory issues required us to split the data collection into several year ranges. Second, the output files sometimes contain multiple rows for the same article, breaking the csv format. The script data_preparation/combine_octoparse_outputs.py fixes these issues by combining the files ajps_article_content_2003_2007.csv, ajps_article_content_2008_2012.csv and ajps_article_content_2013_2016.csv into bld/ajps_articles_2003_2016.csv. Analogously, data_preparation/create_article_info_from_issue_toc.py cleans ajps_issue_toc.csv into bld/ajps_article_info_from_issue_toc.csv.

APSR had a centennial edition which we missed in the first run of data collection (we believe due to a different page layout), and hence collected separately later on. The corresponding file is apsr_article_content_centennial.csv. The script data_preparation/clean_apsr_octoparse_output.py cleans apsr_issue_toc.csv, apsr_article_content_2006_2014.csv and apsr_article_content_centennial.csv and writes them into bld/apsr_article_info_from_issue_toc.csv, bld/apsr_article_content_2006_2014.csv and bld/apsr_centennial_article_content.csv.

Data Entry

There are five categories of data entry.

  • Article: What is the research field of an article, what type of data does it use?
  • Author website: Does an article author's website contain any data or code for the article?
  • Dataverse: Are there any files corresponding to the article on Dataverse?
  • Reference: Does the article content contain a reference to data or code that is useful to reproduce the article's results?
  • Link: Do URLs in the article content that reference code or data, actually lead to such?

The folder data_entry contains an md file describing the protocol on how to enter data for each of these categories. It also contains reference_coding_protocol_questions.md, a collection of clarifying questions on the protocol.

The process for data entry works as follows:

  1. Every combination of journal and data entry category has an ods template.
  2. Several people can work on filling out a template by copying it to their local machine and adding their initials to the file name.
  3. The data entered from several people on the same template is then compared. Differences in entry will be detected across all filled out versions of the same template.
  4. Detected differences are marked in a difference resolution template. Pairs of people with differing data entry meet to resolve differences, and record those in a single copy of the difference resolution file.
  5. The resolved differences are then "harmonized" across difference resolution pairs into a single file.

Each of these steps has corresponding scripts and folders:

  1. Create templates
    • Article: data_preparation/create_article_coding_template.py
    • Author: data_preparation/create_ajps_author_website_coding_template.py, data_preparation/create_apsr_author_website_coding_template.py
    • Dataverse: data_preparation/extract_from_dataverse.py
    • Reference: data_preparation/create_ajps_reference_coding_template.py, data_preparation/create_apsr_reference_coding_template.py, data_preparation/create_econ_reference_coding_template.py
    • Link: data_preparation/create_link_coding_template.py
  2. Upload template. Individual versions of filled out templates are stored and synchronized in URAPShared/Data.
  3. The script data_preparation/create_diff.py can be configured to take multiple individually filled out versions of the same templates and create a single file bld/*_diff.csv listing inconsistencies for each entry across all versions.
  4. The script data_preparation/create_diff_resolution_template.py can be configured to use the bld/*_diff.csv file to create templates for resolution pairs to resolve their differences. These templates are also stored and synchronized in URAPShared/Data, their file names contain the initials of the resolution pair as well as diff_resolution.
  5. The script data_preparation/harmonize.py can be configured to take diff resolution files, and match them to the original templates to create a final bld/*_harmonized.csv file.

The data_preparation folder contains a few more files:

  • import_old_entries.py: This script can be configured to import data entry across different filled out templates. This is useful when changes in the protocol or bug fixes led to additional entries. Existing data entry could be preserved, despite a change in the template file.
  • select_relevant_years.py: Restrict selection of AJPS articles to articles published in years 2006 to 2014. This was necessary because we collected a wider range of articles with Octoparse.
  • update_template_rk.py: This is a script to import old data in response to a very specific change in the template structure.
  • tools.py: A collection of helper functions used across multiple scripts.

CONTENTS:

./analysis ./citation_scraping ./data_cleaning ./data_collection_econ ./data_entry ./data_preparation ./external ./external_econ ./jel_scraping ./logs ./octoparse ./output ./outputforsharelatex ./paper

./analysis

Analysis code for the paper!

./citation_scraping

Scraping code for citations data. We don't use the Selenium-collected data anymore, because the API data is better.

./data_cleaning

Intermediate R and .do files to go from raw data to analysis data. David Birke wrote the code for Poli Sci, Simon Zhu and Neil Tagare built off that for Econ.

./data_collection_econ

Baiyue's scripts to scrape AER and QJE to get the text of the articles.

./data_entry

Protocols for the RAs to do the manual classification parts of the project.

./data_preparation

David Birke's scripts to manage all the raw data and the manual input by the RAs.

./external

A copy of all the files from Box that you need to run the code (Poli Sci)

./external_econ

A copy of all the files from Box that you need to run the code (Econ). Not put into the repo, because Git doesn't store data super well.

./jel_scraping

Scripts to get JEL codes from the ProQuest version of EconLit database.

./logs

Logs from the Stata analysis files in ./Analysis

./octoparse

Octoparse.com files for getting text of poli sci articles.

./output

Every output file produced by the Stata code in ./analysis

./outputforsharelatex

A subset of the Stata code that you manually upload to ShareLaTeX to include in the paper.

./paper

Sorry, not the paper! It's a collection of old slides. Used for presentations.