The hardware and bandwidth for this mirror is donated by METANET, the Webhosting and Full Service-Cloud Provider.
If you wish to report a bug, or if you are interested in having us mirror your free-software or open-source project, please feel free to contact us at mirror[@]metanet.ch.

4. Combine external data sources and COVIDcast data

While the COVIDcast Epidata API provides numerous useful COVID data streams, you may sometimes find yourself with relevant data from a different source. This package provides the tools you need to load such data and use it alongside COVIDcast data—for example, you can calculate correlations with covidcast_cor() and make maps using plot.covidcast_signal().

Let’s illustrate this in action using data from the COVID Tracking Project. They provide CSV files for download containing daily state-level data on numerous metrics, including cases, deaths, hospitalizations, and test results.

Loading the data

We’ve saved a sample of the data in this package to use as an example. Let’s load it and examine a few columns using knitr::kable() to print them nicely:

library(dplyr)

data <- read.csv(system.file("extdata", "covid-tracking-project-oct-2020.csv",
                             package = "covidcast", mustWork = TRUE))

data %>%
    select(date, state, death, deathIncrease, hospitalizedCurrently,
           hospitalizedIncrease) %>%
    head() %>%
    knitr::kable()
date state death deathIncrease hospitalizedCurrently hospitalizedIncrease
2020-10-31 AK 82 1 94 7
2020-10-31 AL 2967 35 960 0
2020-10-31 AR 1925 25 652 35
2020-10-31 AS 0 0 NA 0
2020-10-31 AZ 5979 45 889 68
2020-10-31 CA 17626 55 3212 0

This is in a convenient format: Each day’s observations for each state are in one row. Suppose we would like to extract hospitalizedIncrease as a signal we want to map and analyze alongside other data fetched with covidcast_signal().

To do this, we use as.covidcast_signal(). It expects a data frame with at least three columns:

Other columns are preserved unchanged. (In particular, if an issue column is present, it is used as the issue date of each observation. This is important if your data source includes multiple revisions of each observation.) as.covidcast_signal() also needs to know the geo_type (state, in this case) and source/signal name to apply to the data. With a bit of dplyr data wrangling, we can do this easily:

library(covidcast)

hospitalized <- data %>%
    select(time_value = date,
           geo_value = state,
           value = hospitalizedIncrease) %>%
    mutate(geo_value = tolower(geo_value),
           time_value = as.Date(time_value)) %>%
    as.covidcast_signal(geo_type = "state",
                        data_source = "covid-tracking",
                        signal = "hospitalized_increase")

head(hospitalized) %>%
    knitr::kable()
data_source signal geo_value time_value value issue
covid-tracking hospitalized_increase ak 2020-10-31 7 2023-07-11
covid-tracking hospitalized_increase al 2020-10-31 0 2023-07-11
covid-tracking hospitalized_increase ar 2020-10-31 35 2023-07-11
covid-tracking hospitalized_increase as 2020-10-31 0 2023-07-11
covid-tracking hospitalized_increase az 2020-10-31 68 2023-07-11
covid-tracking hospitalized_increase ca 2020-10-31 0 2023-07-11

This allows us to make maps using the same functions used for other COVIDcast data:

plot(hospitalized, plot_type = "choro")
## Warning: Metadata for signal mean and standard deviation not available;
## defaulting to observed mean and standard deviation to set plot range.

Analysis alongside other signals

Now that our data is loaded as a covidcast_signal object, we can use it alongside other signals from the API. For example, let’s examine how new COVID hospitalizations correlate with outpatient doctor visits with deaths during October 2020, where we use death data as reported by the API.

deaths <- covidcast_signal("jhu-csse", "deaths_7dav_incidence_prop",
                           start_day = "2020-10-01",
                           end_day = "2020-10-31",
                           geo_type = "state")

covidcast_cor(deaths, hospitalized, by = "time_value")
## # A tibble: 31 × 2
##    time_value  value
##    <date>      <dbl>
##  1 2020-10-01 0.218 
##  2 2020-10-02 0.120 
##  3 2020-10-03 0.147 
##  4 2020-10-04 0.148 
##  5 2020-10-05 0.0436
##  6 2020-10-06 0.0706
##  7 2020-10-07 0.123 
##  8 2020-10-08 0.0300
##  9 2020-10-09 0.0955
## 10 2020-10-10 0.178 
## # ℹ 21 more rows

We can also use the tools provided by this package to place both signals into a single data frame for analysis. For example, to build a model that uses hospitalizations and other data to predict deaths, it may be convenient to produce a data frame where each row represents one state on one day, and each column is a variable (such as hospitalization or death). Using aggregate_signals(), this is easy:

death_hosp <- aggregate_signals(list(deaths, hospitalized),
                                format = "wide")

head(death_hosp) %>%
    knitr::kable()
geo_value time_value value+0:jhu-csse_deaths_7dav_incidence_prop value+0:covid-tracking_hospitalized_increase
ca 2020-10-01 0.2286116 0
ca 2020-10-02 0.2137337 0
ca 2020-10-03 0.2122822 0
ca 2020-10-04 0.2122822 0
ca 2020-10-05 0.2177254 0
ca 2020-10-06 0.1788977 0

These binaries (installable software) and packages are in development.
They may not be fully stable and should be used with caution. We make no claims about them.