Inference under Superspreading

This repository provides replication code for the paper “Inference under Superspreading” by Patrick Schmidt.

Installation

You can replicate the paper by downloading all relevant files from GitHub and installing the package. You can use the command

# install.packages("devtools")
devtools::install_github("Schmidtpk/InfSup")

Then, get the R-package containing all data from CovidGer.

Structure

All data for replication is gathered in a separate R-package CovidGer. This includes case data by the Robert Koch Institute, weather data by the German Weather Service (DWD), and data on policy interventions. All data is available on state and county level. The case data contains date of symptom onset, date of reporting, death, and age brackets.

Preparation of data and execution of MCMC sampling can be found in the folder jags models/submission/run model.R. The JAGS model can be found in the same folder and is named jags base error.R.

The replication code of all figures, tables, and quantitative statements is provided in analysis/plots.R.

How to anlyse the data

The MCMC sampling runs several days. The sample MCMC chain along with other information on the model is saved in the repo and can be analyzed directly.

The main plot in paper is generated with the function show_effects_sample, which can average age-specific effect estimates.

library(InfSup)
library(tidyverse)
#> Warning: package 'tidyverse' was built under R version 4.1.3
#> Warning: package 'tibble' was built under R version 4.1.3
#> Warning: package 'tidyr' was built under R version 4.1.3
#> Warning: package 'readr' was built under R version 4.1.3
#> Warning: package 'dplyr' was built under R version 4.1.3
#> Warning: package 'stringr' was built under R version 4.1.3
#> Warning: package 'forcats' was built under R version 4.1.3

#load data and macro information about model
load("./jags models/submission/save/submission.RData")

cov.main<-c("UPM.Relative_Feuchte",
            "TMK.Lufttemperatur",
            "info_lincidence",
            "traced",
            "speechlimited",
            "narrowtesting",
            "shops",
            "sports",
            "gatheringslimited",
            "restaurants",
            "distance",
            "masks",
            "churchesopen",
            "schoolsopenlimited",
            "curfew",
            "education")

# show average effects ordered by size
show_effects_sample(cov.choice = cov.main,
                    average.age = TRUE,
                    order = TRUE)
#> Warning: `position_dodge()` requires non-overlapping x intervals

Alternatively, the results can be shown for all age groups separatedly:

show_effects_sample(cov.choice = cov.main)

For continuous variables, it is interesting to see how the total effect of a covariate changed over time. The function show_total_effect takes the model inputs and computes and plots their effect over time based on the estimates in the model. It averages over all locations if average_c is TRUE and over all age groups if average_age is TRUE.

show_total_effect(average_c = TRUE,
                  average_age = TRUE,
                  covariates = c("traced"))
#> Warning: Duplicated aesthetics after name standardisation: alpha

For the weather variables the function can forecast the total effect based on past weather of the last years.

show_total_effect(covariates = macro$cov.weather,
                  average_age = TRUE,
                  forecast = TRUE,
                  smooth = 14)
#> Warning: Duplicated aesthetics after name standardisation: alpha