The ICEWS event data are one of the major event data sources publicly available, currently providing more than 17 million events from 1995 to late 2018 for more than 198 independent countries. The data were at first delivered with a one-year delay, but since late 2018 are available with monthly updates and no embargo period.

The data are delivered via Dataverse, an open-source research data sharing platform. The main repository at contains documentation and data in the form of 20+ data files covering each year from 1995 to the present. Another repository at has been providing weekly data updates since early 2020.1 Although it is straightforward to bulk-download the entire repository, the data files are contained in a mix of “.tab” and zipped “” files, file versions can change and minimally change every month for the data file for the current year, and the bulk-download itself is several GB in size.

The icews package provides several features to make it easier to work with the ICEWS event data:

  • programmatic download of the event data to a user-specified location that can be persisted between R sessions via relevant “.Rprofile” entries (see setup_icews)
  • keep an updated local copy of the event data, with new downloads or updates in place as needed when new files or file versions are available on dataverse (see update_icews)
  • the option to keep either only the data files, or a SQLite database containing the data, or both
  • data helpers for reading the entire data into memory (read_icews), regardless of the backend used (files or database), read and properly format a single data file (read_events_tsv), or connect (connect) or directly query the database (query_icews)

Setup and data acquisition

The package is setup primarily with the intention that a SQLite database will be used to store the events. A more minimalist, alternative workflow is to work with the raw tab-separated data files only.

At the moment, there isn’t really any advantage to using the database option over working with the raw files. It takes around 2 minutes for my laptop to read the raw data into memory, which itself takes up a bit over 2 GB of space, but once that is done it is easy to work interactively with the data in R. With indices, the database takes up a larger space on disk, more than 8 GB, and the queries themselves should take longer than in-memory work. But for a specific query with the right indices in the database, it is probably faster to use SQL than to load the whole data into memory and then do the query/data transformations.

The most flexible, and recommended option, is to use both files and database. Usually the several GB of extra space this takes up is not a constraint. The next three sections will cover the intial setup and subsequent workflow for the recommended dual, and database or files-only workflows.

Note: the package is inherently unfriendly to a completely replicable workflow, as it is setup with the intention that one will use R options set in .Rprofile at each session start. And in any case, it is not trivial to throw around 5 GB of data.

One side effect is that several parts of this vignette incorporte results from the local data copy I have, but which is not included on GitHub or the package install. However, I have tried to make the code here as explicity and replicable as possible.

Database-only setup

For this workflow, we will setup the options to use a database to store the events, and to discard the raw data files.

setup_icews("~/path/to/icews_data", use_db = TRUE, keep_files = FALSE,
            r_profile = TRUE)

update_icews will now download the “.tab” data files to a temporary file and ingest from there. This saves about 5GB of space as of late 2018.

Files-only setup

Again, we start with the setup function and add the relevant options to the “.Rprofile” file in order to avoid having to deal with paths in future sessions.

setup_icews("~/path/to/icews_data", use_db = FALSE, keep_files = TRUE,
            r_profile = TRUE)

update_icews will now only download and keep in sync the local data files.

A completely minimalistic alternative that does not require options to be set is to use download_data:

old_opts <- unset_icews_opts()
download_data(to_dir = "~/Downloads/icews_data", update = TRUE, dryrun = TRUE)
File system changes:
Found 0 local data file(s)
Downloading 24 file(s)
Removing 0 old file(s)

Download            ''
Download            ''

Data overview

The package includes a data sample for the 4,993 events betwen 1 and 3 June 2018. This provides an overview of the data structure:

#> tibble [4,993 × 22] (S3: tbl_df/tbl/data.frame)
#>  $ event_id       : int [1:4993] 32889836 32889837 32891290 32891065 32891064 32889838 32891059 32891060 32891062 32891061 ...
#>  $ event_date     : Date[1:4993], format: "2018-06-01" "2018-06-01" ...
#>  $ source_name    : chr [1:4993] "Foreign Affairs (Bangladesh)" "High Ranking Military Personnel (Vietnam)" "Ministry (Japan)" "Japan" ...
#>  $ source_sectors : chr [1:4993] "Executive,Foreign Ministry,Government" "Military,Government" "Government" NA ...
#>  $ source_country : chr [1:4993] "Bangladesh" "Vietnam" "Japan" "Japan" ...
#>  $ event_text     : chr [1:4993] "Consult" "Consult" "Consider policy option" "Express intent to engage in diplomatic cooperation (such as policy support)" ...
#>  $ cameo_code     : chr [1:4993] "040" "040" "014" "032" ...
#>  $ intensity      : num [1:4993] 1 1 0 4.5 4.5 0 3 4 2.5 4 ...
#>  $ target_name    : chr [1:4993] "High Ranking Military Personnel (Vietnam)" "Foreign Affairs (Bangladesh)" "Citizen (Japan)" "Vietnam" ...
#>  $ target_sectors : chr [1:4993] "Military,Government" "Executive,Foreign Ministry,Government" "General Population / Civilian / Social,Social" NA ...
#>  $ target_country : chr [1:4993] "Vietnam" "Bangladesh" "Japan" "Vietnam" ...
#>  $ story_id       : int [1:4993] 47400351 47400351 47400357 47400359 47400359 47400360 47400361 47400361 47400361 47400361 ...
#>  $ sentence_number: int [1:4993] 2 2 5 2 2 3 2 3 3 3 ...
#>  $ publisher      : chr [1:4993] "Vietnam News Agency Bulletin" "Vietnam News Agency Bulletin" "Mainichi Daily News" "Vietnam News Agency Bulletin" ...
#>  $ city           : chr [1:4993] "Hanoi" "Hanoi" "Osaka" NA ...
#>  $ district       : chr [1:4993] NA NA "Osaka-shi" NA ...
#>  $ province       : chr [1:4993] "Thanh Pho Ha Noi" "Thanh Pho Ha Noi" "Osaka-fu" NA ...
#>  $ country        : chr [1:4993] "Vietnam" "Vietnam" "Japan" "Japan" ...
#>  $ latitude       : num [1:4993] 21 21 34.7 35.6 21 ...
#>  $ longitude      : num [1:4993] 106 106 136 140 106 ...
#>  $ year           : int [1:4993] 2018 2018 2018 2018 2018 2018 2018 2018 2018 2018 ...
#>  $ yearmonth      : int [1:4993] 201806 201806 201806 201806 201806 201806 201806 201806 201806 201806 ...

The country column includes short country names for a slightly expanded version of the countries and territories with an ISO 3166-1 code.2

Format differences between the raw data and database version

By default when importing data, and always in the database, column names are normalized by converting to lower case and replacing white space with underscore, e.g. “Event ID” becomes “event_id”.

In regard to column data types, the “event_date” column on import is converted to an R Date class vector. SQLite does not have a date data type. There is some limited support for dates in standard character format, e.g. YYYY-MM-DD, but the event dates are instead stored as integers with format YYYYMMDD in the database since the associated index is faster.

Some example queries

The next section below gives a more thorough overview of the ICEWS data; here are some simple example transformations of the data, with several alternative ways to do each.

First, the total number of events per year, with three different methods giving equivalent results:

events1 <- query_icews("SELECT year, count(*) as events
                        FROM events 
                        GROUP BY year;")

# dplyr
events2 <- tbl(connect(), "events") %>%
  group_by(year) %>%
  summarize(events = n()) %>%

# read-in memory then dplyr
events3 <- read_icews() %>%
  group_by(year) %>%
  summarize(events = n())

# plot the country series
ggplot(events1, aes(x = year, y = events)) +
  geom_line() +

Next we will do country-year totals, using the ICEWS country column.

events1 <- query_icews("SELECT country, year, count(*) as events
                        FROM events 
                        GROUP BY year, country;")

# dplyr
events2 <- tbl(connect(), "events") %>%
  group_by(year, country) %>%
  summarize(events = n()) %>%

# read-in memory then dplyr
events3 <- read_icews() %>%
  group_by(year, country) %>%
  summarize(events = n())

# plot the country series
ggplot(events1, aes(x = year, y = events, group = country)) +
  geom_line() +

The ICEWS country names, and the ISO standard they are derived from, for state-centric analysis and data aggregations: (1) they tend to disregard states that do not exist anymore, e.g. Yugoslavia, and state changes in general, and (2) a lot of non-independent territories are treated separately from their associated states, e.g. Puerto Rico, American Samoa, etc. have seperate designations from the US. The Gleditsch and Ward (GW) list of independent states3, which extends and improves upon the Correlates of War (COW) state list4, provides an alternative coding that aims to capture independent states and preserve historical accuracy.

events <- query_icews("SELECT event_date, country FROM events;")
gw_totals <- events %>%
    # convert integer to proper Date
    event_date = as.Date(as.character(event_date), format = "%Y%m%d", 
                         origin = "1970-01-01"),
    # translate country to GW codes
    gwcode = icews_to_gwcode(country, event_date),
    # extract year for aggregation
    year = as.integer(format(event_date, "%Y"))
  ) %>%
  group_by(gwcode, year) %>%
  summarize(events = n())

# plot the GW country series
ggplot(gw_totals, aes(x = year, y = events, group = gwcode)) +
  geom_line() +

  1. Originally there was a repository with daily updates, and the original version of the icews package was setup to work with this. This daily repo stopped updating in the fall of 2019; when updates resumed in 2020, it was replaced by the current repository with weekly updates. If you see references to daily updates anywhere, these are accidental leftovers from the old package version.↩︎

  2. At least it appears to be this.↩︎

  3. See and↩︎

  4. See “State System Membership” at↩︎