Replication archive for Hughes et al. (2021)

Overview

This repository contains replication materials for the paper “Using Administrative Records and Survey Data to Construct Samples of Tweeters and Tweets.”.

This project involved two research groups (Pew and Northeastern) analyzing separate data sets (the samples described in the paper). Because of the terms of our data-sharing agreements, motivated by a desire to protect user privacy as much as possible, these samples were analyzed separately. For this reason, much of the analysis is split across two folders, voter_file_panel and pew_panels.

The code for constructing the voter file sample was originally developed for Grinberg et al. (2019); that code can be found here.

pew_panels

Overview of contents:

• internet_phone_poll.R: analysis of RDD phone poll (Figure 3)
• poq_replication_survey.R: analysis of KP and ATP samples (all figures and Table 1)
• poq_tweetextanalysis.R: analysis of tweet text for KP/ATP (Table 2)

voter_file_panel

Overview of contents:

• definitions.py: Utility file, describing file schemas.
• extract_demographics.py: Pre-processing script for constructing sample demographics
• demographics.ipynb: Analysis notebook, using output of extract_demographics.py (Figures 1 and 2)
• behavior.py: Pre-processing script for constructing sample Tweet metadata
• behavior.ipynb: Analysis notebook, using output of behavior.py (Figure 4)
• summarize_for_poq_neu_pew.R: Script for combining histogram data (Table 1, Figure 4)
• histograms.sh: Script for computing voter-file histograms (Table 1)
• hash/: Construction of sample intersections (Appendix A5)
• hash_agreement.ipynb: Analysis of sample intersections (Appendix A5)