Facebook Survey Dashboard

Sampling

library(covidcast)
library(dplyr)
library(ggplot2)

df_fb1_st = covidcast_signal("fb-survey", "raw_cli", geo_type = "state")
n_per_day = df_fb1_st %>% 
  group_by(time_value) %>% 
  summarize(n = sum(sample_size)) 

ggplot(n_per_day, aes(x = time_value, y = n)) + 
  geom_line() + geom_point() + theme_bw() + 
  labs(x = "Date", y = "Number of Responses",
       title = sprintf("Total responses: %i, mean per day: %i", 
                       round(sum(n_per_day$n)), round(mean(n_per_day$n))))

df_fb1 = covidcast_signal("fb-survey", "smoothed_cli")
county_per_week = df_fb1 %>% 
  group_by(time_value) %>%
  summarize(n = n())

ggplot(county_per_week, aes(x = time_value, y = n)) + 
  geom_line() + geom_point() + theme_bw() + 
  labs(x = "Date", y = "Number of Counties", 
       title = sprintf("Unique counties: %i, mean per week: %i", 
                       length(unique(df_fb1$geo_value)), 
                       round(mean(county_per_week$n))))

Coverage

county_totals = df_fb1 %>% 
  group_by(geo_value) %>%
  summarize(value = n()) %>% ungroup() %>%
  filter(substr(geo_value,3,5) != "000") 

county_totals$time_value = "2020-04-15"
county_totals$issue = "2020-04-15"
attributes(county_totals)$geo_type = "county"
class(county_totals) = c("covidcast_signal", "data.frame")

plot(county_totals, title = "Which counties ever appear in our CLI signals?",
     choro_col = c("#D3D3D3", "#FFC0CB"), choro_params = list(breaks = c(0, 1), 
                                                              legend_width = 5))

Correlation

df_fb2 = covidcast_signal("fb-survey", "smoothed_hh_cmnty_cli")
df_in = covidcast_signal("jhu-csse", "confirmed_7dav_incidence_prop", 
                         start_day = min(df_fb1$time_value))

case_num = 500
geo_values = covidcast_signal("jhu-csse", "confirmed_cumulative_num",
                              max(df_in$time_value), max(df_in$time_value)) %>%
  filter(value >= case_num) %>% pull(geo_value)
df_fb1_act = df_fb1 %>% filter(geo_value %in% geo_values)
df_fb2_act = df_fb2 %>% filter(geo_value %in% geo_values)
df_in_act = df_in %>% filter(geo_value %in% geo_values)

df_cor1 = covidcast_cor(df_fb1_act, df_in_act, by = "time_value",
                        method = "spearman")
df_cor2 = covidcast_cor(df_fb2_act, df_in_act, by = "time_value",
                        method = "spearman")

df_cor = rbind(df_cor1, df_cor2)
df_cor$signal = c(rep("% CLI", nrow(df_cor1)), 
                  rep("% CLI-in-community", nrow(df_cor2)))
ggplot(df_cor, aes(x = time_value, y = value)) + 
  geom_line(aes(color = signal)) + 
  labs(title = "Correlation between CLI signals and cases",
       subtitle = sprintf("Over all counties with at least %i cases", case_num),
       x = "Date", y = "Correlation") + theme_bw() +
  theme(legend.position = "bottom", legend.title  = element_blank())