Rebuild Notes

A Note on version 1.0.0 and later

The bulk of the data included in troopdata versions under v1.0.0 were drawn from the original Kane data, which was collected from 1950-2005. The data were then updated through 2020 by Michael A. Allen, Michael E. Flynn, and Carla Martinez Machain. The data were then further updated through 2022 by Michael Flynn. Version 1.0.0 marked a total rebuild of the troopdata package and so many values may not match up exactly with the original Kane data. Here are a few key points to note regarding the updates made since v1.0.0:

• The older DoD reports differentiate between forces ashore and forces afloat, but often have certain afloat forces assigned to a particular country. The data keep the two apart: the country figures (troops_ad, navy_ad, marine_corps_ad) are personnel ashore, and the afloat personnel a report assigns to a country are held beside them. The afloat argument of get_troopdata() decides what to do with them. afloat = "exclude", the default, leaves them out of the country totals; afloat = "include" adds them in; and afloat = "separate" returns them in their own columns (troops_afloat, navy_afloat and marine_corps_afloat) next to the ashore figures. Omitting afloat forces attached to a particular country can severely underestimate the number of personnel present. In Greece in 1974, for example, we see 3,800 personnel ashore and a total of 5,437 reported by the DoD as assigned to Greece, including afloat: "exclude" returns the first figure and "include" the second. Note that for 1953 through 1967 the reports assign afloat personnel to the country of the nearest port, so "include" can place a fleet at sea in whichever country it was closest to on the day of the count (Greece reads 15,066 for 1960 with 1,976 ashore).

• The original reports also contain afloat personnel that are not assigned to particular countries, such as the “other afloat” category and, in more recent years, an afloat line for each region that is not disaggregated by country. These are not aggregated into any country totals. They are carried in a separate location named “Afloat” (country code 10200), which covers 1950 and 1953 through 2007. With afloat = "exclude" that location holds everyone afloat worldwide; with "include" or "separate" it holds only those not assigned to a country, so the worldwide total is the same either way. The one afloat line the reports give for the United States and its territories together is treated as the United States’ own. From September 2008 the DoD counts a ship’s crew at its home port, inside the country or state figure, and there is no separate afloat figure.

• For more recent years some discrepancies between versions might be the result of the fact that the DoD issued multiple reports with different figures, even for the same monthly release. For example, There are multiple reports issued for September 2008 with each reporting different country-specific values.

• From December 2015 through December 2017 the reports list a large number of personnel, nearly all of them Navy, inside the United States block but in no state: about 77,000 under “Unknown” in December 2015, and between 88,500 and 104,703 in each report from March 2016 to December 2017 under “Armed Forces Europe”, “Armed Forces Pacific” and “Armed Forces the Americas”. Because they cannot be traced to a state or a country they are not included in any of the figures, so the United States total and its Navy figure are lower in those quarters and return in March 2018. The data are left as the reports print them.

• There are also some smaller territorial units that appear in the DoD reports that did not appear in Kane’s original data or our package. We create ad hoc country codes to deal with them. Hong Kong, for example, is treated separately in rebuilt data. This can also factor into higher regional totals whre Kane’s data may have omitted some of these smaller territorial units.

• In general we try to assign country codes to deployments that correspond to their actual physical geographic location. For example, deployments to Greenland could be assigned to Denmark, but we assign them to Greenland with an ad hoc country code. Similarly, deployments to Puerto Rico are assigned to Puerto Rico rather than the United States. This is a change from the original Kane data, which assigned all deployments to the United States. These values may be aggregated by users for analysis if they wish to do so.

• In more recent years the DoD has issued multiple quarterly reports in a given year. When reporting annual values the get_troopdata() function will return the highest value from the quarterly reports for a given year where those are available. This should only affect more recent years, but it’s a different approach than Kane’s original method which involved just using the September reports. The next section goes through this rule and the others like it.

• In some cases the physical location of a deployment may remain constant but the polity associated with that deployment may change. For example, the force in Kosovo is counted under Yugoslavia from 1999 through 2005, under Serbia in 2006 and 2007, and under Kosovo from 2008.

• The DoD often lists China and Hong Kong as a single entity. Prior to 1997 deployments to Hong Kong were usually listed separately, but following 1997 they are often grouped together. We adopt an ad hoc country code for Hong Kong and assign deployments to that code where they are listed as such in the DoD reports. After 1997 we treat any deployment references to Hong Kong as deployments to China.

How values are handled

This section lays out the rules that turn the DoD reports into the values that get_troopdata() returns. The help page for get_troopdata() has the same information in shorter form.

Annual values are the highest value reported in the year

The data are stored by report period. There is one report per year through 2012 (dated June for 1950 and 1953 through 1956, and September from 1957), two for 2013 (September and December), and four per year from 2014 (March, June, September and December). The most recent year only has the reports published so far, so 2026 currently rests on the March report alone.

When you ask for annual data, which is the default, each column holds the largest value reported for that location in that year. A few things follow from this:

• It is not an average of the quarters, and it is not the value from a fixed month like September.
• The largest value is found for each column separately. The total can come from one report and the Army or Navy figure from another. For Japan in 2019 the four reports show 56,134, 55,327, 55,245 and 57,094 active duty personnel. The annual troops_ad value is 57,094, from December. But the annual army_ad value (2,671) is from June and the annual navy_ad value (20,846) is from March.
• Because of this, the branch columns in an annual row do not have to add up to troops_ad, and troops_all does not have to equal troops_ad plus the guard and reserve columns.
• If you want figures that all come from the same report, use quarters = TRUE. In the quarterly country data troops_ad always equals the sum of the branch columns for a row that comes from a report.

Totals

troops_ad is active duty personnel. troops_all is troops_ad plus the seven National Guard and Reserve columns, and you get it by setting guard_reserve = TRUE. The reports only give guard and reserve figures from 2008, so before that troops_all is the same as troops_ad.

The United States figure is the fifty states and the District of Columbia. Puerto Rico, Guam, the US Virgin Islands and the other territories are separate locations with their own codes and are not part of the United States figure.

Zeros and missing values

A country that a report does not list gets a 0 for that period. So a zero can mean that the report shows zero, or that the country is not in the report at all.

Territories and the other locations with our ad hoc codes work differently, and so do a few small states in the years before their independence. They only have a row for the periods in which a report lists them. If you ask for Bermuda for 1995 through 2006, for example, you get no rows back rather than a set of zeros.

NA means the reports do not give that kind of figure. That covers the guard, reserve and civilian columns and the Coast Guard before 2008, the Space Force before September 2023, the branch columns for 1951 and 1952, and the afloat columns for the years and places where the reports do not assign afloat personnel to a location. In the annual data a value is only NA if it is missing in every report of that year.

Values that do not come from a report

Most values are exactly what a report shows. These are the exceptions. The source column in the troopdata_rebuild_long data frame identifies most of them. The two it does not flag are the Army quarters of 2022 and 2023, which carry the name of their report, and the Iraq figure for 2018 through 2020, which has no source label.

• 1951 and 1952. There are no reports for these two years. The total for each country moves in equal steps from the June 1950 figure to the June 1953 figure, and the branch columns are NA. Most territories, and the Afloat location, have no rows at all for these two years.
• December 2022, March 2023 and June 2023. The Army did not provide personnel data for these three quarters while it converted to a new personnel system. In the country data the Army’s active duty, National Guard and Reserve figures move in equal steps from the September 2022 report to the September 2023 report, and the totals are those figures plus what the other services reported. Everything the other services reported is left as published. The data for US states and the report data (reports = TRUE) are not filled in, so the Army figures and the totals are NA for those quarters. (The United States row in the report data is the one exception. It is added up from the states and carries a fixed Army figure for those quarters.)
• June 2023. The June 2023 workbook has two blocks of its overseas list printed one row too low: the civilian columns from Montenegro to the end of the list, and five of the guard and reserve columns from Morocco to Wake Island. Qatar’s row shows Puerto Rico’s 2,201 civilians, and Uruguay’s shows the United Kingdom’s 1,383. The neighboring reports and the report’s own totals confirm where the figures belong, so the country data read them from the row below the one they are printed on. The report data (reports = TRUE) keep the sheet as published.
• War zones. The reports do not show the personnel deployed to Iraq, Kuwait and Afghanistan for much of the 2000s, and from December 2017 they leave out deployed personnel in Afghanistan, Iraq and Syria. For some of those years we use figures from Kane’s data or from public reporting: Afghanistan for 2001-2005 and 2018-2020, Iraq for 2003-2007 and 2018-2021, Kuwait for 2003-2007, and Syria for 2018-2021. These are totals only, so the branch columns are zero for those rows. We do not have outside figures after those years. The values for Afghanistan from 2021 and for Iraq and Syria from 2022 are the small numbers the reports show, and they are well below the actual US presence. In the quarterly data the outside figures for Iraq in 2018-2021 and for Syria in 2021 are on the June row only. The other quarters of those years show what the report prints, which is zero or close to it, so use the annual data for these cases.
• Rows dated June between 1957 and 2013. The reports for those years are dated September (and December in 2013). If you pull quarterly data for those years you will also see some rows dated June. These hold figures from Kane’s data, or one of the estimates above, and they are kept only for a location and year where no report gives a figure, or where the report prints a zero in its place. Nearly all of them are zero.

Where personnel are counted

• Personnel are counted in the location the report lists them under. We do not move them to a parent country, so Greenland is Greenland and not Denmark.
• Personnel that a report lists under no location are not in the data. The older reports have lines such as “Transients”, “Undistributed” and “Departmental Headquarters”, and the newer ones have “Unknown”. These can be large (about 207,000 people in 1968), so adding up every row in the data gives you less than the worldwide total printed in the report.
• When a report lists two places under what we treat as one country, the country figure is the sum. Japan and the Ryukyu Islands (Okinawa) are the main case, through 1973. Setting reports = TRUE shows each of those rows separately.
• For 1950 through 2007 the country figures are personnel ashore, and the afloat argument controls what happens to personnel afloat (see above). From September 2008 the reports count a ship’s crew at its home port.
• From December 2015 to December 2017 the reports list between 77,000 and 105,000 personnel, nearly all Navy, in the United States block but in no state. They are not in any of the figures.

Country codes

The numeric codes are Gleditsch and Ward codes, not Correlates of War codes. The two lists mostly agree, but Germany is 260 here (the Federal Republic, before and after 1990) and East Germany is 265. The two Congos are 484, the Republic of the Congo (Brazzaville), and 490, the Democratic Republic of the Congo (Kinshasa, formerly Zaire). The reports of 1960 through 1962 have a single “Congo” line, which is the former Belgian Congo and is counted under 490. Locations that the Gleditsch and Ward list does not cover have ad hoc codes, for example 6 for Puerto Rico, 1002 for Greenland, 1004 for Diego Garcia, 1008 for Guam and 10200 for personnel afloat. The construction data from get_builddata() use the same codes for the same places. The basing data from get_basedata() have not been rebuilt yet and still use the older codes, so Germany is 255 there.