Download the COVID-19 case counts for England and fit a simple smoothed model accounting for day-of-the-week

covid_eng_fit.R

library(dplyr)
library(mgcv)
library(ggplot2)

# Download
c19 <- readr::read_csv("https://coronavirus.data.gov.uk/downloads/csv/coronavirus-cases_latest.csv")

# Extract case counts for England and select useful variables
c19_Eng <- c19 %>% filter(`Area type` == "nation") %>%
  transmute(date = `Specimen date`, cases = `Daily lab-confirmed cases`)

# Limit to date range with positive number of cases and prepare for regression
c19_Eng_reg <- c19_Eng %>% 
  filter(date > as.Date("2020-02-22") & date < as.Date("2020-10-01")) %>%
  mutate(weekday = weekdays(date), day_count = as.integer(date - as.Date("2020-02-22")))

# Generalised additive model of cases as a smooth function of time (since 2020-02-22) and accounting for a day-of-the-week effect
c19.gam <- gam(cases ~ s(day_count) + weekday - 1, data = c19_Eng_reg, subset = date < as.Date("2020-09-25"), family = nb)

# Add the fitted line to the data
c19_Eng_reg <- c19_Eng_reg %>% mutate(fitted = NA)
c19_Eng_reg$fitted[c19_Eng_reg$date < as.Date("2020-09-25")] <- c19.gam$fitted.values

# Plot the points and the line
print(
  ggplot(c19_Eng_reg, aes(x = date, y = cases)) + geom_point() + geom_line(aes(x = date, y = fitted)) + scale_y_log10()
)