This directory has the code needed to replicate the results in "Quantifying population contact patterns in the United States during the COVID-19 pandemic"

• code/ - has the code used in the analysis
• data/ - [will be created by a script] has the survey data used in the analysis (NOTE: because of its size, the data are not included in the git repo; they will be downloaded by the script 00-run-all.r)
• out/ - [will be created by a script] where the results of the scripts get saved (this is also not included in the git repo; it gets created by the scripts)

DATA

We provide data from the surveys we conducted. To replicate our comparison with the contact matrix from Prem et al (2017), you will need to follow the instructions below.

The data we provide are

• data/
  • df_all_waves.rds - the survey responses
  • df_boot_all_waves.rds - bootstrap weights to accompany df_all_waves.rds
  • df_alters_all_waves.rds - file with info about detailed contacts reported by respondentts
  • df_alters_boot_all_waves.rds - bootstrap weights to accompany df_alters_all_waves.rds
• data/ACS
  • acs15_fb_agecat_withkids.rds - total number of people in the US in 2015 by age categories used in FB survey
  • acs15_wave0_agecat.rds - total number of people in the US in 2015 by age cateogries used in our survey
  • acs18_national_targets.rds - distribution of 2018 US population by covariates that we use in calibration weighting
  • acs18_prem_agecat.rds - distribution of 2018 US population by age groups that match Prem et al (2017)
  • acs18_wave0_agecat.rds - total number of people in the US in 2018 by age categories used in our survey
  • acs18_wave1_agecat_withkids.rds - total number of people in the US in 2018 by age categories that include children
• data/fb-2015-svy
  • fb_ego.csv - survey responses
  • fb_alters.csv - detailed contacts reported in survey
  • fb_bootstrapped_weights.csv - bootstrap weights to accompany fb_ego.csv
• data/prem_contact_matrix
  • prem_usa.csv - estimated contact matrix for the United States from Prem et al. (2017). NOTE: This data must be downloaded by you. See instructions below.
• data/polymod
  • [this directory starts empty, but has files generated in it by the scripts]
• data/contact-matrices
  • [this directory starts empty, but has files generated in it by the scripts]

Data from Prem et al. (2017) - You must download this data from here. The contact matrix for the US (for all locations) is available in a tab in the MUestimates_all_locations_2.xlsx file in the downloaded folder. Save this tab as a csv file with the name prem_usa.csv in the prem_contact_matrix subfolder in the data folder.

NOTE: The first time you run 00-run-all.R (see below), the last script (35-sensitivity-compare-with-Prem-matrix.Rmd) will stop with an error because it can't find the prem_usa.csv datafile. Once you create this file following the above instructions, you will be able to run 00-run-all.R without any errors.

CODE

These scripts were run on a 2020 Macbook Pro with 8 cores and 64 GB of memory. Because of the large number of bootstrap resamples, some of the files take time to run. We try to give a rough sense for expected runtime below by providing estimated runtimes for files that take longer than 5 minutes.

• 00-run-all.R - this file downloads the data and runs all of the scripts
• 01-prep-wave-comparison-model - this prepares a dataset that is later used in the models
• 10-compare-waves - this file analyzes the sample composition and marginal distributions of number of contacts, relationship, and location of contacts
• 11-relationships - this file calculates the share of reported contacts by wave and relationship
• 12-locations - this file calculates the average number of reported contacts by wave and location
• 13-figure - this file creates the figure out of the histograms, relationships, and locations
• 14-sensitivity - [about 10 minutes to run] this file assesses the sensitivity of results to including / not including physical contact in Waves 1 and 2
• 15-avg-contacts - this file fits a model to estimate the average number of contacts by wave
• 12-model_mean_perwave - this model estimates average number of contacts, accounting for censoring, using a model
• 21-allcc_nb_censored_loaded_weighted - this is the model for non-household contacts (with covariates)
• 22-nonhhcc_nb_censored_loaded_weighted - this is the model for all contacts (with covariates)
• 23-plot_model_predictions_by_covars - this file produces the plots that show model predictions
• 30-contact-matrices - [about 2 hours to run] this file calculates the contact matrices
• 31-prep-estimate-R0-bootstrap - this prepares the dataset that is used for the epidemiological analyses
• 32-estimate-R0 - this file generates age-structured contact matrices from the survey data and the corresponding R₀ estimates
• 33-sensitivity-estimate-R0-onlycc - this file assesses the sensitivity of the R₀ estimates to including / not including physical contact in Waves 1 and 2
• 34-sensitivity-high-low-baselineR0 - this file assesses the sensitivity of the R₀ estimates to assuming higher and lower baseline values
• 35-sensitivity-compare-with-Prem-matrix.Rmd - this file compares the data from Feehan and Cobb (2019) with estimates from Prem et al. (2017) and the UK POLYMOD data from Mossong et al. (2008)

Additionally, there are two files that have some miscellaneous helper functions:

• model-coef-plot-helpers.R
• utils.R

DOCKER

It is likely that you have different versions of R and specific R packages than we did when we wrote our code. Thus, we recommend using Docker to replicate our results. Using Docker will ensure that you have exactly the same computing environment that we did when we conducted our analyses.

To use Docker

1. Install Docker Desktop (if you don't already have it)
2. Clone this repository
3. Be sure that your current working directory is the one that you downloaded the repository into. It's probably called bics-paper-code/
4. Build the docker image. docker build --rm -t bics-replication . This step will likely take a little time, as Docker builds your image (including installing various R packages)
5. Run the docker image docker run -d --rm -p 8888:8787 -e PASSWORD=pass --name bics bics-replication
6. Open a web browser and point it to localhost:8888
7. Log onto Rstudio with username 'rstudio' and password 'pass'
8. Open the file bics-paper-code/code/00-run-all.r
9. Running the file should replicate everything. If you have not downloaded the Prem et al (2017) data, the last script (35-sensitivity-compare-with-Prem-matrix.Rmd) will stop with an error because it can't find the prem_usa.csv datafile.