Paper replication package and source code.
This repository contains three components
- Paper replication package COVIDPoliticalEvents.jl
- Code to conduct the paper analyses (see "analysis")
- Code to reconstruct the data used in the paper (see "analysis/data")
This directory contains the Julia code used to replicate main and supplemental findings in the paper "Local COVID-19 mortality did not increase after large-scale political events in the USA in 2020 and 2021".
Files that must be downloaded by the user are linked, files that are included in the package are noted, and appear in the "data" directory.
- JHU COVID (deaths, cases)
- County level estimates of R_t (from covidestim)
- US Census / ACS (via tidycensus)
- BLS unemployment ("bls_unemployment.csv")
- US Census region definitions ("census_delineation.csv")
- US county adjacency information
- US Urban-rural code classification ("ruralurbancodes2013.csv)
- NYT county-level masking
- ALCED protest events ("final_protest_data.csv")
- Primary elections turnout ("2020_presidential_primary_turnout.csv")
- GA election turnout ("ga_election_results_clean.csv")
- SafeGraph mobility
- Download the covidestim county-level data, and the mobility data (see above), into the "data" directory.
- Specify your Census Data API key, in "preprocess.R", as an argument to
preprocess(). - If you possess the SafeGraph mobility data, you must specify it as "patpth" as an argument to
process_csv()in "preprocess.jl". N.B., this data is not freely available, but is provided by SafeGraph, Inc. - Execute "covid_data_make.sh" to generate "cvd_dat.jld2", used for the main analyses.
TSCSMethods works on a standard computer, with sufficient RAM and processing power to support the size of the dataset analyzed by the user. This will be a computer with at least 16 GB, and 4 cores. Analysis and testing was carried out on a system running MAC OS 17.0, with 64 Gb RAM, and an intel i9 processor @ 2.30Ghz.
While analysis was executed on a MAC OSX system, all of the underlying dependencies are compatible with Windows, Mac, and Linux systems. This package has been tested on Julia 1.7.1. Data was additionally processed using R 4.1.
R system information and platform details
_
platform x86_64-apple-darwin17.0
arch x86_64
os darwin17.0
system x86_64, darwin17.0
status
major 4
minor 1.0
year 2021
month 05
day 18
svn rev 80317
language R
version.string R version 4.1.0 (2021-05-18)
nickname Camp Pontanezen
Julia system information and platform details
Julia Version 1.7.1
Commit ac5cc99908 (2021-12-22 19:35 UTC)
Platform Info:
OS: macOS (x86_64-apple-darwin21.1.0)
CPU: Intel(R) Core(TM) i9-9880H CPU @ 2.30GHz
WORD_SIZE: 64
LIBM: libopenlibm
LLVM: libLLVM-13.0.0 (ORCJIT, skylake)
Environment:
JULIA_EDITOR = code
JULIA_NUM_THREADS = 16
- R version 4.1.0 (2021-05-18)
- Packages: dplyr, magrittr, tibble, lubridate, ggplot2 (each available through the CRAN repository)
- Julia version 1.7.1 (2021-12-22)
- Packages: TSCSMethods, COVIDPoliticalEvents, Random, DataFrames, Dates, CSV, JLD2 (all except the first two available through Julia's package manager)
All packages are available through the standard repositories, except TSCSMethods and COVIDPoliticalEvents.
Julia and R may be installed on Mac OSX using homebrew by executing:
brew install juliabrew install rOtherwise, consult the Julia Language download page and the R Language download page for installation on your system.
From within a julia session, type:
import Pkg; Pkg.add("https://github.com/human-nature-lab/TSCSMethods.jl")
import Pkg; Pkg.add("https://github.com/human-nature-lab/COVIDPoliticalEvents.jl")
The packages should take approximately 1-2 minutes to install.
After installation of the software dependencies, a given model should take anywhere from 10-60 minutes to execute, depending on workstation specifications.
The models for a given scenario may be executed together, by executing the "run" file, e.g., "run_blm.jl", or separately, by executing the file for a specific model (e.g., "base_model.jl"). Each scenario is housed in a specific subdirectory.
The output for each model is saved as a Julia Data Format file (".jld2"), and the file structure itself depends on the above packages.
Executing a "run" file creates output files for each model in the "out", subdirectories for each scenario.
You may inspect model output according to the following:
using TSCSMethods, JLD2
output = load_object("ga full_death_rte_.jld2")
the output object contains the following fields:
- model (without refinement)
- refinedmodel (refined to best matches)
- calmodel (with caliper)
- refcalmodel (refined to best matches, with a caliper)
- matchinfo (information about the matches)
- obsinfo (info about the treated observations)
For example, access the results of the refined caliper model as:
output.refcalmodel.results
Which yields the ATT estimates, confidence intervals, and the number of treated units from a given model.
The latter two fields contain information about the matched units and the treated observation units.
For a simple program example, to estimate the ATTs for a specific event, in a simple context, see "ga-election/base_model.jl" which runs through estimation of the overall ATTs for the Georgia special election, each scenario has the same overall structure:
The functions to generate the main, extended data, and supporting figures (and their dependencies) are contained directly within this package. The scripts to generate the figure sets are in analysis/plotting.
N.B. these figures depend on model output files.